Datasets:

---

JetBrains-Research

/

commit-chronicle

like 9

Follow

JetBrains Research 71

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNNConvolutionAlgorithm.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNNConvolutionAlgorithm.java", "diff": "@@ -217,29 +217,32 @@ public class LibMatrixCuDNNConvolutionAlgorithm implements java.lang.AutoCloseab\npublic static LibMatrixCuDNNConvolutionAlgorithm cudnnGetConvolutionBackwardDataAlgorithm(\nGPUContext gCtx, String instName, int N, int C, int H, int W, int K, int R, int S,\nint pad_h, int pad_w, int stride_h, int stride_w, int P, int Q, long workspaceLimit) throws DMLRuntimeException {\n- //long t1 = DMLScript.FINEGRAINED_STATISTICS ? System.nanoTime() : 0;\nLibMatrixCuDNNConvolutionAlgorithm ret = new LibMatrixCuDNNConvolutionAlgorithm(gCtx, instName, N, C, H, W, K, R, S,\npad_h, pad_w, stride_h, stride_w, P, Q);\n-\n+ if(H == R || W == S) {\n// CuDNN's cudnnGetConvolutionBackwardDataAlgorithm returns CUDNN_CONVOLUTION_BWD_DATA_ALGO_1 for atleast one scenario\n// for sentence CNN (N=1, C=1, H=2060, W=300, F=500, Hf=5, Wf=300, sparsity=0.1).\n// This causes more than 100x slowdown when compared with CUDNN_CONVOLUTION_BWD_DATA_ALGO_0.\n- // To keep things simple for now, we will always prefer to use memory-less operator: CUDNN_CONVOLUTION_BWD_DATA_ALGO_0\n+ // To keep things simple for now, we will always prefer to use memory-less operator for conv1d: CUDNN_CONVOLUTION_BWD_DATA_ALGO_0\nret.algo = jcuda.jcudnn.cudnnConvolutionBwdDataAlgo.CUDNN_CONVOLUTION_BWD_DATA_ALGO_0;\n-// int[] algos = {-1};\n-// long sizeInBytesArray[] = {Math.min(workspaceLimit, MAX_WORKSPACE_LIMIT_BYTES)};\n-// jcuda.jcudnn.JCudnn.cudnnGetConvolutionBackwardDataAlgorithm(\n-// LibMatrixCuDNN.getCudnnHandle(gCtx),\n-// ret.filterDesc, ret.nkpqTensorDesc, ret.convDesc, ret.nchwTensorDesc,\n-// cudnnConvolutionBwdDataPreference.CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT, sizeInBytesArray[0], algos);\n-// jcuda.jcudnn.JCudnn.cudnnGetConvolutionBackwardDataWorkspaceSize(LibMatrixCuDNN.getCudnnHandle(gCtx),\n-// ret.filterDesc, ret.nkpqTensorDesc, ret.convDesc, ret.nchwTensorDesc, algos[0], sizeInBytesArray);\n-// if (sizeInBytesArray[0] != 0)\n-// ret.workSpace = gCtx.allocate(sizeInBytesArray[0]);\n-// ret.sizeInBytes = sizeInBytesArray[0];\n-// ret.algo = algos[0];\n-// if (DMLScript.FINEGRAINED_STATISTICS)\n-// GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_CUDNN_INIT, System.nanoTime() - t1);\n+ }\n+ else {\n+ long t1 = DMLScript.FINEGRAINED_STATISTICS ? System.nanoTime() : 0;\n+ int[] algos = {-1};\n+ long sizeInBytesArray[] = {Math.min(workspaceLimit, MAX_WORKSPACE_LIMIT_BYTES)};\n+ jcuda.jcudnn.JCudnn.cudnnGetConvolutionBackwardDataAlgorithm(\n+ LibMatrixCuDNN.getCudnnHandle(gCtx),\n+ ret.filterDesc, ret.nkpqTensorDesc, ret.convDesc, ret.nchwTensorDesc,\n+ jcuda.jcudnn.cudnnConvolutionBwdDataPreference.CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT, sizeInBytesArray[0], algos);\n+ jcuda.jcudnn.JCudnn.cudnnGetConvolutionBackwardDataWorkspaceSize(LibMatrixCuDNN.getCudnnHandle(gCtx),\n+ ret.filterDesc, ret.nkpqTensorDesc, ret.convDesc, ret.nchwTensorDesc, algos[0], sizeInBytesArray);\n+ if (sizeInBytesArray[0] != 0)\n+ ret.workSpace = gCtx.allocate(sizeInBytesArray[0]);\n+ ret.sizeInBytes = sizeInBytesArray[0];\n+ ret.algo = algos[0];\n+ if (DMLScript.FINEGRAINED_STATISTICS)\n+ GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_CUDNN_INIT, System.nanoTime() - t1);\n+ }\nreturn ret;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/JCudaKernels.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/JCudaKernels.java", "diff": "@@ -27,6 +27,7 @@ import java.io.IOException;\nimport java.io.InputStream;\nimport java.util.HashMap;\n+import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.io.IOUtilFunctions;\nimport org.apache.sysml.runtime.matrix.data.LibMatrixCUDA;\n@@ -110,6 +111,7 @@ public class JCudaKernels {\ncheckResult(cuLaunchKernel(function, config.gridDimX, config.gridDimY, config.gridDimZ, config.blockDimX,\nconfig.blockDimY, config.blockDimZ, config.sharedMemBytes, config.stream, Pointer.to(kernelParams),\nnull));\n+ if(DMLScript.SYNCHRONIZE_GPU)\nJCuda.cudaDeviceSynchronize();\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/AggUnaryOp.java", "new_path": "src/main/java/org/apache/sysml/hops/AggUnaryOp.java", "diff": "@@ -551,6 +551,10 @@ public class AggUnaryOp extends Hop implements MultiThreadedHop\n{\nboolean ret = false;\n+ // TODO: Disable ternary aggregate rewrite on GPU backend.\n+ if(DMLScript.USE_ACCELERATOR)\n+ return false;\n+\n//currently we support only sum over binary multiply but potentially\n//it can be generalized to any RC aggregate over two common binary operations\nif( OptimizerUtils.ALLOW_SUM_PRODUCT_REWRITES && _op == AggOp.SUM &&\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/mllearn/estimators.py", "new_path": "src/main/python/systemml/mllearn/estimators.py", "diff": "@@ -896,7 +896,7 @@ class Keras2DML(Caffe2DML):\n\"\"\"\n- def __init__(self, sparkSession, keras_model, input_shape, transferUsingDF=False, weights=None, labels=None, batch_size=64, max_iter=2000, test_iter=10, test_interval=500, display=100, lr_policy=\"step\", weight_decay=5e-4, regularization_type=\"L2\"):\n+ def __init__(self, sparkSession, keras_model, input_shape, transferUsingDF=False, load_keras_weights=True, weights=None, labels=None, batch_size=64, max_iter=2000, test_iter=10, test_interval=500, display=100, lr_policy=\"step\", weight_decay=5e-4, regularization_type=\"L2\"):\n\"\"\"\nPerforms training/prediction for a given keras model.\n@@ -906,6 +906,7 @@ class Keras2DML(Caffe2DML):\nkeras_model: keras model\ninput_shape: 3-element list (number of channels, input height, input width)\ntransferUsingDF: whether to pass the input dataset via PySpark DataFrame (default: False)\n+ load_keras_weights: whether to load weights from the keras_model. If False, the weights will be initialized to random value using NN libraries' init method (default: True)\nweights: directory whether learned weights are stored (default: None)\nlabels: file containing mapping between index and string labels (default: None)\nbatch_size: size of the input batch (default: 64)\n@@ -931,6 +932,7 @@ class Keras2DML(Caffe2DML):\nconvertKerasToCaffeNetwork(keras_model, self.name + \".proto\", int(batch_size))\nconvertKerasToCaffeSolver(keras_model, self.name + \".proto\", self.name + \"_solver.proto\", int(max_iter), int(test_iter), int(test_interval), int(display), lr_policy, weight_decay, regularization_type)\nself.weights = tempfile.mkdtemp() if weights is None else weights\n+ if load_keras_weights:\nconvertKerasToSystemMLModel(sparkSession, keras_model, self.weights)\nif labels is not None and (labels.startswith('https:') or labels.startswith('http:')):\nimport urllib\n@@ -939,6 +941,7 @@ class Keras2DML(Caffe2DML):\nfrom shutil import copyfile\ncopyfile(labels, os.path.join(weights, 'labels.txt'))\nsuper(Keras2DML,self).__init__(sparkSession, self.name + \"_solver.proto\", input_shape, transferUsingDF)\n+ if load_keras_weights:\nself.load(self.weights)\ndef close(self):\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cpp/CMakeLists.txt", "new_path": "src/main/cpp/CMakeLists.txt", "diff": "@@ -25,7 +25,7 @@ set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} \"${CMAKE_SOURCE_DIR}/cmake/\")\n# Options to Use OpenBLAS or Intel MKL\noption(USE_OPEN_BLAS \"Whether to use OpenBLAS (Defaults to compiling with Intel MKL, if both set, MKL has priority)\" OFF)\n-option(USE_INTEL_MKL \"Whether to use Intel MKL (Defaults to compiling with Intel MKL)\" ON)\n+option(USE_INTEL_MKL \"Whether to use Intel MKL (Defaults to compiling with Intel MKL)\" OFF)\n# Build a shared libraray\nadd_library(systemml SHARED libmatrixdnn.cpp libmatrixmult.cpp systemml.cpp)\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so", "new_path": "src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so", "diff": "Binary files a/src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so and b/src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so differ\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so", "new_path": "src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so", "diff": "Binary files a/src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so and b/src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so differ\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixdnn.cpp", "new_path": "src/main/cpp/libmatrixdnn.cpp", "diff": "@@ -480,6 +480,7 @@ int sconv2dBiasAddDense(float* inputPtr, float* biasPtr, float* filterPtr, float\n// Step 3: Destroy the description of the operation\ndnnDelete_F32(pConvolution);\n+ return computeNNZ<float>(retPtr, N*KPQ);\n#else\n// First step: Avoids oversubscription and other openmp/internal blas threading issues\nsetNumThreadsForBLAS(1);\n@@ -515,7 +516,7 @@ int sconv2dBiasAddDense(float* inputPtr, float* biasPtr, float* filterPtr, float\nnnz += computeNNZ<float>(retPtr + n*KPQ, KPQ);\n}\ndelete [] loweredMatArrays;\n+ return nnz;\n#endif\n- return nnz;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixmult.cpp", "new_path": "src/main/cpp/libmatrixmult.cpp", "diff": "#include <cstdlib>\n#include \"omp.h\"\n#include <cmath>\n+\n+#ifdef USE_OPEN_BLAS\n#include <cblas.h>\n+#endif\nint SYSML_CURRENT_NUM_THREADS = -1;\nvoid setNumThreadsForBLAS(int numThreads) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixmult.h", "new_path": "src/main/cpp/libmatrixmult.h", "diff": "// we call \"extension\" APIs for setting number of threads of the given API.\n// For example: for OpenBLAS we use openblas_set_num_threads and\n// for MKL we use mkl_set_num_threads. This avoids performance degradation due to overprovisioning.\n-#ifdef USE_OPEN_BLAS\n-#include <cblas.h>\n-extern \"C\" void openblas_set_num_threads(int numThreads);\n-#elif defined USE_INTEL_MKL\n+#ifdef USE_INTEL_MKL\n#include <mkl.h>\n#include <mkl_service.h>\n+#else\n+ #include <cblas.h>\n+ extern \"C\" void openblas_set_num_threads(int numThreads);\n#endif\nvoid setNumThreadsForBLAS(int numThreads);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so", "new_path": "src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so", "diff": "Binary files a/src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so and b/src/main/cpp/lib/libsystemml_mkl-Linux-x86_64.so differ\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so", "new_path": "src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so", "diff": "Binary files a/src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so and b/src/main/cpp/lib/libsystemml_openblas-Linux-x86_64.so differ\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixdnn.cpp", "new_path": "src/main/cpp/libmatrixdnn.cpp", "diff": "@@ -406,8 +406,8 @@ int dconv2dBiasAddDense(double* inputPtr, double* biasPtr, double* filterPtr, do\n// Step 3: Destroy the description of the operation\ndnnDelete_F64(pConvolution);\n+ return computeNNZ<double>(retPtr, N*KPQ);\n#else\n- // ------------------------------------------------------------------------------------\n// First step: Avoids oversubscription and other openmp/internal blas threading issues\nsetNumThreadsForBLAS(1);\n@@ -418,8 +418,9 @@ int dconv2dBiasAddDense(double* inputPtr, double* biasPtr, double* filterPtr, do\n// Allocate temporary data structures used in parallel for\nint numOpenMPThreads = MIN(numThreads, N);\ndouble* loweredMatArrays = new double[numIm2ColElem*numOpenMPThreads];\n+ int nnz = 0;\n-#pragma omp parallel for num_threads(numOpenMPThreads)\n+#pragma omp parallel for reduction(+: nnz) num_threads(numOpenMPThreads)\nfor (int n = 0; n < N; n++) {\nint threadID = omp_get_thread_num();\ndouble* loweredMat = loweredMatArrays + numIm2ColElem*threadID;\n@@ -436,12 +437,13 @@ int dconv2dBiasAddDense(double* inputPtr, double* biasPtr, double* filterPtr, do\ndouble* outputArr = retPtr + n*KPQ;\nif( addBias )\nbiasAdd<double>(biasPtr, outputArr, K, PQ);\n- } // end omp parallel for\n+\n+ // Step 4: thread-local nnz maintenance\n+ nnz += computeNNZ<double>(retPtr + n*KPQ, KPQ);\n+ }\ndelete [] loweredMatArrays;\n- // ------------------------------------------------------------------------------------\n+ return nnz;\n#endif\n-\n- return computeNNZ<double>(retPtr, N*KPQ);\n}\nint sconv2dBiasAddDense(float* inputPtr, float* biasPtr, float* filterPtr, float* retPtr,\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixmult.cpp", "new_path": "src/main/cpp/libmatrixmult.cpp", "diff": "#ifdef USE_OPEN_BLAS\n#include <cblas.h>\n+#else\n+ #include <mkl_service.h>\n#endif\nint SYSML_CURRENT_NUM_THREADS = -1;\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/libmatrixmult.h", "new_path": "src/main/cpp/libmatrixmult.h", "diff": "//#endif\n// Since we call cblas_dgemm in openmp for loop,\n-// we call \"extension\" APIs for setting number of threads of the given API.\n-// For example: for OpenBLAS we use openblas_set_num_threads and\n-// for MKL we use mkl_set_num_threads. This avoids performance degradation due to overprovisioning.\n+// we call \"extension\" APIs for setting the number of threads.\n#ifdef USE_INTEL_MKL\n#include <mkl.h>\n#include <mkl_service.h>\n+ extern \"C\" void mkl_set_num_threads(int numThreads);\n#else\n#include <cblas.h>\nextern \"C\" void openblas_set_num_threads(int numThreads);\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/systemml.cpp", "new_path": "src/main/cpp/systemml.cpp", "diff": "@@ -75,9 +75,10 @@ JNIEXPORT void JNICALL Java_org_apache_sysml_utils_NativeHelper_setMaxNumThreads\nmaxThreads = (int) jmaxThreads;\n}\n-JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_matrixMultDenseDense(\n+JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_dmmdd(\nJNIEnv* env, jclass cls, jdoubleArray m1, jdoubleArray m2, jdoubleArray ret,\n- jint m1rlen, jint m1clen, jint m2clen, jint numThreads) {\n+ jint m1rlen, jint m1clen, jint m2clen, jint numThreads)\n+{\ndouble* m1Ptr = GET_DOUBLE_ARRAY(env, m1, numThreads);\ndouble* m2Ptr = GET_DOUBLE_ARRAY(env, m2, numThreads);\ndouble* retPtr = GET_DOUBLE_ARRAY(env, ret, numThreads);\n@@ -92,6 +93,21 @@ JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_matrixMultDe\nreturn (jboolean) true;\n}\n+JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_smmdd(\n+ JNIEnv* env, jclass cls, jobject m1, jobject m2, jobject ret,\n+ jint m1rlen, jint m1clen, jint m2clen, jint numThreads)\n+{\n+ float* m1Ptr = (float*) env->GetDirectBufferAddress(m1);\n+ float* m2Ptr = (float*) env->GetDirectBufferAddress(m2);\n+ float* retPtr = (float*) env->GetDirectBufferAddress(ret);\n+ if(m1Ptr == NULL || m2Ptr == NULL || retPtr == NULL)\n+ return (jboolean) false;\n+\n+ smatmult(m1Ptr, m2Ptr, retPtr, (int)m1rlen, (int)m1clen, (int)m2clen, (int)numThreads);\n+\n+ return (jboolean) true;\n+}\n+\nJNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_tsmm\n(JNIEnv * env, jclass cls, jdoubleArray m1, jdoubleArray ret, jint m1rlen, jint m1clen, jboolean isLeftTranspose, jint numThreads) {\ndouble* m1Ptr = GET_DOUBLE_ARRAY(env, m1, numThreads);\n" }, { "change_type": "MODIFY", "old_path": "src/main/cpp/systemml.h", "new_path": "src/main/cpp/systemml.h", "diff": "@@ -28,12 +28,18 @@ extern \"C\" {\n#endif\n/*\n* Class: org_apache_sysml_utils_NativeHelper\n- * Method: matrixMultDenseDense\n- * Signature: ([D[D[DIIII)Z\n+ * Method: dmmdd\n*/\n-JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_matrixMultDenseDense\n+JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_dmmdd\n(JNIEnv *, jclass, jdoubleArray, jdoubleArray, jdoubleArray, jint, jint, jint, jint);\n+/*\n+ * Class: org_apache_sysml_utils_NativeHelper\n+ * Method: smmdd\n+ */\n+JNIEXPORT jboolean JNICALL Java_org_apache_sysml_utils_NativeHelper_smmdd\n+ (JNIEnv *, jclass, jobject, jobject, jobject, jint, jint, jint, jint);\n+\n/*\n* Class: org_apache_sysml_utils_NativeHelper\n* Method: tsmm\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNHelper.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNHelper.java", "diff": "@@ -68,7 +68,7 @@ public class LibMatrixDNNHelper\nret.sparse = false;\nif(ret.getDenseBlock() == null)\nret.allocateDenseBlock();\n- NativeHelper.matrixMultDenseDense(m1.getDenseBlockValues(), m2.getDenseBlockValues(),\n+ NativeHelper.dmmdd(m1.getDenseBlockValues(), m2.getDenseBlockValues(),\nret.getDenseBlockValues(), m1.rlen, m1.clen, m2.clen, 1);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixNative.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixNative.java", "diff": "@@ -34,7 +34,10 @@ import org.apache.sysml.utils.Statistics;\npublic class LibMatrixNative\n{\n- /** ThreadLocal reuse of direct buffers for inputs/outputs (extended on demand).*/\n+ // ThreadLocal reuse of direct buffers for inputs/outputs (extended on demand).\n+ // note: since we anyway have to convert from double to float, we use\n+ // preallocated direct buffers (with thread-local reuse and resizing on demand)\n+ // to ensure there are no additional copies created by the transfer over jni\nprivate static ThreadLocal<FloatBuffer> inBuff = new ThreadLocal<FloatBuffer>();\nprivate static ThreadLocal<FloatBuffer> biasBuff = new ThreadLocal<FloatBuffer>();\nprivate static ThreadLocal<FloatBuffer> filterBuff = new ThreadLocal<FloatBuffer>();\n@@ -65,32 +68,45 @@ public class LibMatrixNative\nk = k <= 0 ? NativeHelper.getMaxNumThreads() : k;\n// check inputs / outputs\n- if (m1.isEmptyBlock() || m2.isEmptyBlock()) {\n+ if (m1.isEmptyBlock(false) || m2.isEmptyBlock(false)){\nret.setNonZeros(0);\nif(examSparsity)\nret.examSparsity(); // turn empty dense into sparse\nreturn;\n}\n- if (NativeHelper.isNativeLibraryLoaded() &&\n- !isMatMultMemoryBound(m1.rlen, m1.clen, m2.clen) && !m1.isInSparseFormat() && !m2.isInSparseFormat()) {\n+\n+ if (NativeHelper.isNativeLibraryLoaded()\n+ && !isMatMultMemoryBound(m1.rlen, m1.clen, m2.clen)\n+ && !m1.isInSparseFormat() && !m2.isInSparseFormat())\n+ {\nret.sparse = false;\nret.allocateDenseBlock();\nlong start = DMLScript.STATISTICS ? System.nanoTime() : 0;\n- if (NativeHelper.matrixMultDenseDense(m1.getDenseBlockValues(), m2.getDenseBlockValues(),\n- ret.getDenseBlockValues(), m1.getNumRows(), m1.getNumColumns(), m2.getNumColumns(), k)) {\n+ boolean rccode = false;\n+ if( isSinglePrecision() ) {\n+ FloatBuffer fin1 = toFloatBuffer(m1.getDenseBlockValues(), inBuff, true);\n+ FloatBuffer fin2 = toFloatBuffer(m2.getDenseBlockValues(), filterBuff, true);\n+ FloatBuffer fout = toFloatBuffer(ret.getDenseBlockValues(), outBuff, false);\n+ rccode = NativeHelper.smmdd(fin1, fin2, fout,\n+ m1.getNumRows(), m1.getNumColumns(), m2.getNumColumns(), k);\n+ fromFloatBuffer(outBuff.get(), ret.getDenseBlockValues());\n+ }\n+ else {\n+ rccode = NativeHelper.dmmdd(m1.getDenseBlockValues(), m2.getDenseBlockValues(),\n+ ret.getDenseBlockValues(), m1.getNumRows(), m1.getNumColumns(), m2.getNumColumns(), k);\n+ }\n+ if (rccode) {\nif(DMLScript.STATISTICS) {\nStatistics.nativeLibMatrixMultTime += System.nanoTime() - start;\nStatistics.numNativeLibMatrixMultCalls.increment();\n}\nret.recomputeNonZeros();\n- // post-processing (nnz maintained in parallel)\nif(examSparsity)\nret.examSparsity();\nreturn;\n- } else {\n- // Else fall back to Java\n- Statistics.incrementNativeFailuresCounter();\n}\n+ //else record failure and fallback to java\n+ Statistics.incrementNativeFailuresCounter();\n}\nif (k == 1)\nLibMatrixMult.matrixMult(m1, m2, ret, examSparsity);\n@@ -135,14 +151,9 @@ public class LibMatrixNative\nelse {\nif(params.bias.isInSparseFormat())\nparams.bias.sparseToDense(); // Bias matrix is usually extremely small\n- boolean singlePrecision = ConfigurationManager.getDMLConfig()\n- .getTextValue(DMLConfig.FLOATING_POINT_PRECISION).equals(\"single\");\nlong start = DMLScript.STATISTICS ? System.nanoTime() : 0;\nint nnz = -1;\n- if( singlePrecision ) {\n- //note: since we anyway have to convert from double to float, we use\n- //preallocated direct buffers (with thread-local reuse and resizing on demand)\n- //to ensure there are no additional copies created by the transfer over jni\n+ if( isSinglePrecision() ) {\nFloatBuffer finput = toFloatBuffer(input.getDenseBlockValues(), inBuff, true);\nFloatBuffer fbias = toFloatBuffer(params.bias.getDenseBlockValues(), biasBuff, true);\nFloatBuffer ffilter = toFloatBuffer(filter.getDenseBlockValues(), filterBuff, true);\n@@ -260,6 +271,11 @@ public class LibMatrixNative\nLibMatrixDNN.conv2dBackwardData(filter, dout, outputBlock, params);\n}\n+ private static boolean isSinglePrecision() {\n+ return ConfigurationManager.getDMLConfig()\n+ .getTextValue(DMLConfig.FLOATING_POINT_PRECISION).equals(\"single\");\n+ }\n+\nprivate static FloatBuffer toFloatBuffer(double[] input, ThreadLocal<FloatBuffer> buff, boolean copy) {\n//maintain thread-local buffer (resized on demand)\nFloatBuffer ret = buff.get();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/NativeHelper.java", "new_path": "src/main/java/org/apache/sysml/utils/NativeHelper.java", "diff": "@@ -324,7 +324,12 @@ public class NativeHelper {\n}\n// TODO: Add pmm, wsloss, mmchain, etc.\n- public static native boolean matrixMultDenseDense(double [] m1, double [] m2, double [] ret, int m1rlen, int m1clen, int m2clen, int numThreads);\n+\n+ //double-precision matrix multiply dense-dense\n+ public static native boolean dmmdd(double [] m1, double [] m2, double [] ret, int m1rlen, int m1clen, int m2clen, int numThreads);\n+ //single-precision matrix multiply dense-dense\n+ public static native boolean smmdd(FloatBuffer m1, FloatBuffer m2, FloatBuffer ret, int m1rlen, int m1clen, int m2clen, int numThreads);\n+ //transpose-self matrix multiply\nprivate static native boolean tsmm(double [] m1, double [] ret, int m1rlen, int m1clen, boolean isLeftTranspose, int numThreads);\n// ----------------------------------------------------------------------------------------------------------------\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "diff": "@@ -35,20 +35,22 @@ public class CNodeBinary extends CNode\nVECT_POW_ADD, VECT_MIN_ADD, VECT_MAX_ADD,\nVECT_EQUAL_ADD, VECT_NOTEQUAL_ADD, VECT_LESS_ADD,\nVECT_LESSEQUAL_ADD, VECT_GREATER_ADD, VECT_GREATEREQUAL_ADD,\n- VECT_CBIND_ADD,\n+ VECT_CBIND_ADD, VECT_XOR_ADD,\n//vector-scalar operations\nVECT_MULT_SCALAR, VECT_DIV_SCALAR, VECT_MINUS_SCALAR, VECT_PLUS_SCALAR,\nVECT_POW_SCALAR, VECT_MIN_SCALAR, VECT_MAX_SCALAR,\nVECT_EQUAL_SCALAR, VECT_NOTEQUAL_SCALAR, VECT_LESS_SCALAR,\nVECT_LESSEQUAL_SCALAR, VECT_GREATER_SCALAR, VECT_GREATEREQUAL_SCALAR,\nVECT_CBIND,\n+ VECT_XOR_SCALAR,\n//vector-vector operations\nVECT_MULT, VECT_DIV, VECT_MINUS, VECT_PLUS, VECT_MIN, VECT_MAX, VECT_EQUAL,\nVECT_NOTEQUAL, VECT_LESS, VECT_LESSEQUAL, VECT_GREATER, VECT_GREATEREQUAL,\n+ VECT_XOR,\n//scalar-scalar operations\nMULT, DIV, PLUS, MINUS, MODULUS, INTDIV,\nLESS, LESSEQUAL, GREATER, GREATEREQUAL, EQUAL,NOTEQUAL,\n- MIN, MAX, AND, OR, LOG, LOG_NZ, POW,\n+ MIN, MAX, AND, OR, XOR, LOG, LOG_NZ, POW,\nMINUS1_MULT, MINUS_NZ;\npublic static boolean contains(String value) {\n@@ -88,6 +90,7 @@ public class CNodeBinary extends CNode\ncase VECT_MINUS_ADD:\ncase VECT_PLUS_ADD:\ncase VECT_POW_ADD:\n+ case VECT_XOR_ADD:\ncase VECT_MIN_ADD:\ncase VECT_MAX_ADD:\ncase VECT_EQUAL_ADD:\n@@ -112,6 +115,7 @@ public class CNodeBinary extends CNode\ncase VECT_MINUS_SCALAR:\ncase VECT_PLUS_SCALAR:\ncase VECT_POW_SCALAR:\n+ case VECT_XOR_SCALAR:\ncase VECT_MIN_SCALAR:\ncase VECT_MAX_SCALAR:\ncase VECT_EQUAL_SCALAR:\n@@ -142,6 +146,7 @@ public class CNodeBinary extends CNode\ncase VECT_DIV:\ncase VECT_MINUS:\ncase VECT_PLUS:\n+ case VECT_XOR:\ncase VECT_MIN:\ncase VECT_MAX:\ncase VECT_EQUAL:\n@@ -199,6 +204,8 @@ public class CNodeBinary extends CNode\nreturn \" double %TMP% = 1 - %IN1% * %IN2%;\\n\";\ncase MINUS_NZ:\nreturn \" double %TMP% = (%IN1% != 0) ? %IN1% - %IN2% : 0;\\n\";\n+ case XOR:\n+ return \" double %TMP% = ( (%IN1% != 0) != (%IN2% != 0) ) ? 1 : 0;\\n\";\ndefault:\nthrow new RuntimeException(\"Invalid binary type: \"+this.toString());\n@@ -217,7 +224,8 @@ public class CNodeBinary extends CNode\n|| this == VECT_EQUAL_SCALAR || this == VECT_NOTEQUAL_SCALAR\n|| this == VECT_LESS_SCALAR || this == VECT_LESSEQUAL_SCALAR\n|| this == VECT_GREATER_SCALAR || this == VECT_GREATEREQUAL_SCALAR\n- || this == VECT_CBIND;\n+ || this == VECT_CBIND\n+ || this == VECT_XOR_SCALAR;\n}\npublic boolean isVectorVectorPrimitive() {\nreturn this == VECT_DIV || this == VECT_MULT\n@@ -225,7 +233,8 @@ public class CNodeBinary extends CNode\n|| this == VECT_MIN || this == VECT_MAX\n|| this == VECT_EQUAL || this == VECT_NOTEQUAL\n|| this == VECT_LESS || this == VECT_LESSEQUAL\n- || this == VECT_GREATER || this == VECT_GREATEREQUAL;\n+ || this == VECT_GREATER || this == VECT_GREATEREQUAL\n+ || this == VECT_XOR;\n}\npublic boolean isVectorMatrixPrimitive() {\nreturn this == VECT_MATRIXMULT\n@@ -351,6 +360,7 @@ public class CNodeBinary extends CNode\ncase VECT_DIV_SCALAR: return \"b(vd)\";\ncase VECT_MINUS_SCALAR: return \"b(vmi)\";\ncase VECT_PLUS_SCALAR: return \"b(vp)\";\n+ case VECT_XOR_SCALAR: return \"v(vxor)\";\ncase VECT_POW_SCALAR: return \"b(vpow)\";\ncase VECT_MIN_SCALAR: return \"b(vmin)\";\ncase VECT_MAX_SCALAR: return \"b(vmax)\";\n@@ -364,6 +374,7 @@ public class CNodeBinary extends CNode\ncase VECT_DIV: return \"b(v2d)\";\ncase VECT_MINUS: return \"b(v2mi)\";\ncase VECT_PLUS: return \"b(v2p)\";\n+ case VECT_XOR: return \"b(v2xor)\";\ncase VECT_MIN: return \"b(v2min)\";\ncase VECT_MAX: return \"b(v2max)\";\ncase VECT_EQUAL: return \"b(v2eq)\";\n@@ -388,6 +399,7 @@ public class CNodeBinary extends CNode\ncase NOTEQUAL: return \"b(!=)\";\ncase OR: return \"b(|)\";\ncase AND: return \"b(&)\";\n+ case XOR: return \"b(xor)\";\ncase MINUS1_MULT: return \"b(1-*)\";\ncase MINUS_NZ: return \"b(-nz)\";\ndefault: return \"b(\"+_type.name().toLowerCase()+\")\";\n@@ -413,6 +425,7 @@ public class CNodeBinary extends CNode\ncase VECT_GREATER_ADD:\ncase VECT_GREATEREQUAL_ADD:\ncase VECT_CBIND_ADD:\n+ case VECT_XOR_ADD:\nboolean vectorScalar = _inputs.get(1).getDataType()==DataType.SCALAR;\n_rows = _inputs.get(vectorScalar ? 0 : 1)._rows;\n_cols = _inputs.get(vectorScalar ? 0 : 1)._cols;\n@@ -435,6 +448,7 @@ public class CNodeBinary extends CNode\ncase VECT_MULT_SCALAR:\ncase VECT_MINUS_SCALAR:\ncase VECT_PLUS_SCALAR:\n+ case VECT_XOR_SCALAR:\ncase VECT_POW_SCALAR:\ncase VECT_MIN_SCALAR:\ncase VECT_MAX_SCALAR:\n@@ -449,6 +463,7 @@ public class CNodeBinary extends CNode\ncase VECT_MULT:\ncase VECT_MINUS:\ncase VECT_PLUS:\n+ case VECT_XOR:\ncase VECT_MIN:\ncase VECT_MAX:\ncase VECT_EQUAL:\n@@ -492,6 +507,7 @@ public class CNodeBinary extends CNode\ncase MAX:\ncase AND:\ncase OR:\n+ case XOR:\ncase LOG:\ncase LOG_NZ:\ncase POW:\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -65,7 +65,7 @@ public class TemplateRow extends TemplateBase\nOpOp1.SIN, OpOp1.COS, OpOp1.TAN, OpOp1.ASIN, OpOp1.ACOS, OpOp1.ATAN, OpOp1.SINH, OpOp1.COSH, OpOp1.TANH,\nOpOp1.CUMSUM, OpOp1.CUMMIN, OpOp1.CUMMAX};\nprivate static final Hop.OpOp2[] SUPPORTED_VECT_BINARY = new OpOp2[]{\n- OpOp2.MULT, OpOp2.DIV, OpOp2.MINUS, OpOp2.PLUS, OpOp2.POW, OpOp2.MIN, OpOp2.MAX,\n+ OpOp2.MULT, OpOp2.DIV, OpOp2.MINUS, OpOp2.PLUS, OpOp2.POW, OpOp2.MIN, OpOp2.MAX, OpOp2.XOR,\nOpOp2.EQUAL, OpOp2.NOTEQUAL, OpOp2.LESS, OpOp2.LESSEQUAL, OpOp2.GREATER, OpOp2.GREATEREQUAL};\npublic TemplateRow() {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "diff": "@@ -561,6 +561,95 @@ public class LibSpoofPrimitives\nreturn vectPlusWrite(b, a, bix, bi, ai, blen, len);\n}\n+ //custom vector xor\n+ /**\n+ * Computes c = xor(A,B)\n+ *\n+ * @param a dense input vector A\n+ * @param ai start position in A\n+ * @param bval scalar value\n+ * @param c resultant vector\n+ * @param ci index of c\n+ * @param len number of processed elements\n+ * @return resultant value\n+ */\n+ public static void vectXorAdd(double[] a, double bval, double[] c, int ai, int ci, int len) {\n+ for( int j = ai; j < ai+len; j++, ci++)\n+ c[ci] += ( (a[j] != 0) != (bval != 0) ) ? 1 : 0;\n+ }\n+\n+ public static void vectXorAdd(double bval, double[] a, double[] c, int ai, int ci, int len) {\n+ for( int j = ai; j < ai+len; j++, ci++)\n+ c[ci] += ( (bval != 0) != (a[j] != 0) ) ? 1 : 0;\n+ }\n+\n+ public static void vectXorAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[ci + aix[j]] += ( (a[j] != 0) != (bval != 0) ) ? 1 : 0;\n+ }\n+\n+ public static void vectXorAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[ci + aix[j]] += ( (bval != 0) != (a[j] != 0) ) ? 1 : 0;\n+ }\n+\n+ //1. scalar vs. dense vector\n+ public static double[] vectXorWrite(double[] a, double bval, int ai, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++)\n+ c[j] = ( ( a[ai+j] != 0) != (bval != 0) ) ? 1 : 0;\n+ return c;\n+ }\n+\n+ //2. dense vector vs. scalar\n+ public static double[] vectXorWrite(double bval, double[] a, int ai, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++)\n+ c[j] = ( (bval != 0) != (a[ai + j] != 0) ) ? 1 : 0;\n+ return c;\n+ }\n+\n+ //3. dense vector vs. dense vector\n+ public static double[] vectXorWrite(double[] a, double[] b, int ai, int bi, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++)\n+ c[j] = ( (a[ai + j] != 0) != (b[bi + j] != 0) ) ? 1 : 0;\n+ return c;\n+ }\n+\n+ //4. sparse vector vs scalar\n+ public static double[] vectXorWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) {\n+ double init = (bval != 0) ? 1 : 0;\n+ double[] c = allocVector(len, true, init);\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[aix[j]] = (a[j] != 0) ? 0 : 1;\n+ return c;\n+ }\n+\n+ //5. scalar vs. sparse vector\n+ public static double[] vectXorWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) {\n+ double init = (bval != 0) ? 1 : 0;\n+ double[] c = allocVector(len, true, init);\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[aix[j]] = (a[j] != 0) ? 0 : 1;\n+ return c;\n+ }\n+\n+ //6. sparse vector vs. dense vector\n+ public static double[] vectXorWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++ )\n+ c[j] = (b[bi+j] != 0) ? 1 : 0;\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[aix[j]] = ( ( a[j] != 0) != (c[aix[j]] != 0) )? 1 : 0;\n+ return c;\n+ }\n+\n+ //6. sparse vector vs. dense vector\n+ public static void vectXorWrite(double[] a, double[] b, int ai, int[] aix, int bi, int alen, int len) {\n+ vectXorWrite(a, b, aix, ai, bi, alen, len);\n+ }\n+\n//custom vector pow\npublic static void vectPowAdd(double[] a, double bval, double[] c, int ai, int ci, int len) {\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CPlanVectorPrimitivesTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CPlanVectorPrimitivesTest.java", "diff": "@@ -645,6 +645,36 @@ public class CPlanVectorPrimitivesTest extends AutomatedTestBase\ntestVectorBinaryPrimitive(BinType.VECT_GREATEREQUAL, InputType.VECTOR_SPARSE, InputType.VECTOR_DENSE);\n}\n+ @Test\n+ public void testScalarVectorXorDense() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR_SCALAR, InputType.SCALAR, InputType.VECTOR_DENSE);\n+ }\n+\n+ @Test\n+ public void testVectorScalarXorDense() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR_SCALAR, InputType.VECTOR_DENSE, InputType.VECTOR_DENSE);\n+ }\n+\n+ @Test\n+ public void testVectorVectorDenseDense() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR, InputType.VECTOR_DENSE, InputType.VECTOR_DENSE);\n+ }\n+\n+ @Test\n+ public void testVectorScalarSparse() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR_SCALAR, InputType.VECTOR_SPARSE, InputType.SCALAR);\n+ }\n+\n+ @Test\n+ public void testScalarVectorSparse() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR_SCALAR, InputType.SCALAR, InputType.VECTOR_SPARSE);\n+ }\n+\n+ @Test\n+ public void testVectorVectorSparseDense() {\n+ testVectorBinaryPrimitive(BinType.VECT_XOR, InputType.VECTOR_SPARSE, InputType.VECTOR_DENSE);\n+ }\n+\n@SuppressWarnings(\"incomplete-switch\")\nprivate static void testVectorAggPrimitive(UnaryType aggtype, InputType type1)\n{\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "diff": "@@ -52,6 +52,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME14 = TEST_NAME+14; //-2 * X + t(Y); t(Y) is rowvector\nprivate static final String TEST_NAME15 = TEST_NAME+15; //colMins(2*log(X))\nprivate static final String TEST_NAME16 = TEST_NAME+16; //colSums(2*log(X));\n+ private static final String TEST_NAME17 = TEST_NAME+17; //xor operation\nprivate static final String TEST_DIR = \"functions/codegen/\";\n@@ -65,7 +66,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for( int i=1; i<=16; i++ ) {\n+ for( int i=1; i<=17; i++ ) {\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(\nTEST_CLASS_DIR, TEST_NAME+i, new String[] {String.valueOf(i)}) );\n}\n@@ -288,6 +289,21 @@ public class CellwiseTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME16, true, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenCellwiseRewrite17() {\n+ testCodegenIntegration( TEST_NAME17, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwise17() {\n+ testCodegenIntegration( TEST_NAME17, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwiseRewrite17_sp() {\n+ testCodegenIntegration( TEST_NAME17, true, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\n@@ -352,6 +368,8 @@ public class CellwiseTmplTest extends AutomatedTestBase\nAssert.assertTrue(!heavyHittersContainsSubString(\"uacmin\"));\nelse if( testname.equals(TEST_NAME16) )\nAssert.assertTrue(!heavyHittersContainsSubString(\"uack+\"));\n+ else if( testname.equals(TEST_NAME17) )\n+ Assert.assertTrue(!heavyHittersContainsSubString(\"xor\"));\n}\nfinally {\nrtplatform = platformOld;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "diff": "@@ -72,6 +72,7 @@ public class RowAggTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME33 = TEST_NAME+\"33\"; //Kmeans, inner loop\nprivate static final String TEST_NAME34 = TEST_NAME+\"34\"; //X / rowSums(X!=0)\nprivate static final String TEST_NAME35 = TEST_NAME+\"35\"; //cbind(X/rowSums(X), Y, Z)\n+ private static final String TEST_NAME36 = TEST_NAME+\"36\"; //xor operation\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + RowAggTmplTest.class.getSimpleName() + \"/\";\n@@ -83,7 +84,7 @@ public class RowAggTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for(int i=1; i<=35; i++)\n+ for(int i=1; i<=36; i++)\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME+i, new String[] { String.valueOf(i) }) );\n}\n@@ -612,6 +613,20 @@ public class RowAggTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME35, false, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenRowAggRewrite36CP() {\n+ testCodegenIntegration( TEST_NAME36, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg36CP() {\n+ testCodegenIntegration( TEST_NAME36, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg36SP() {\n+ testCodegenIntegration( TEST_NAME36, false, ExecType.SPARK );\n+ }\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\n@@ -667,6 +682,8 @@ public class RowAggTmplTest extends AutomatedTestBase\nif( testname.equals(TEST_NAME35) )\nAssert.assertTrue(!heavyHittersContainsSubString(\"spoofRA\", 2)\n&& !heavyHittersContainsSubString(\"cbind\"));\n+ if( testname.equals(TEST_NAME36) )\n+ Assert.assertTrue(!heavyHittersContainsSubString(\"xor\"));\n}\nfinally {\nrtplatform = platformOld;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl17.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+\n+X = matrix(seq(7, 1100*200+6), 1100, 200, byrow=TRUE);\n+\n+R1 = (X/3) %% 0.6;\n+R2 = (X/3) %/% 0.6;\n+\n+R = xor(R1, R2);\n+\n+writeMM(as(R,\"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl17.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(7, 1100*200+6), 1100, 200);\n+\n+R1 = (X/3) %% 0.6;\n+R2 = (X/3) %/% 0.6;\n+\n+S = xor(R1, R2);\n+\n+write(S, $1)\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern36.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+library(\"matrixStats\")\n+\n+X = matrix(seq(1, 6000)/600, 300, 20);\n+\n+Y = X/rowSums(X)\n+\n+R = xor(X,Y);\n+\n+writeMM(as(R, \"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern36.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(1, 6000)/600, 300, 20);\n+\n+Y = X/rowSums(X)\n+\n+S = xor(X, Y);\n+\n+write(S, $1);\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "diff": "@@ -646,8 +646,8 @@ public class LibSpoofPrimitives\n}\n//6. sparse vector vs. dense vector\n- public static void vectXorWrite(double[] a, double[] b, int ai, int[] aix, int bi, int alen, int len) {\n- vectXorWrite(a, b, aix, ai, bi, alen, len);\n+ public static double[] vectXorWrite(double[] a, double[] b, int ai, int[] aix, int bi, int alen, int len) {\n+ return vectXorWrite(a, b, aix, ai, bi, alen, len);\n}\n//custom vector pow\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNRelu.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNRelu.java", "diff": "@@ -75,7 +75,7 @@ public class LibMatrixDNNRelu\nelse if(m1.isInSparseFormat() && !m2.isInSparseFormat())\nreluBackwardSparseDense(m1.getSparseBlock(), m2.getDenseBlock(), out.getSparseBlock(), _rl, _ru);\nelse //sparse-sparse\n- reluBackwardSparseDense(m1.getSparseBlock(), m2.getSparseBlock(), out.getSparseBlock(), _rl, _ru);\n+ reluBackwardSparseSparse(m1.getSparseBlock(), m2.getSparseBlock(), out.getSparseBlock(), _rl, _ru);\n//thread-local nnz maintenance\nreturn out.recomputeNonZeros(_rl, _ru-1);\n@@ -118,7 +118,7 @@ public class LibMatrixDNNRelu\n}\n}\n- private static void reluBackwardSparseDense(SparseBlock a, SparseBlock b, SparseBlock c, int rl, int ru) {\n+ private static void reluBackwardSparseSparse(SparseBlock a, SparseBlock b, SparseBlock c, int rl, int ru) {\n//b is the driver as it has likely less non-zeros\nfor(int i = rl; i < ru; i++) {\nif( a.isEmpty(i) || b.isEmpty(i) ) continue;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -33,7 +33,6 @@ import org.apache.sysml.hops.LiteralOp;\nimport org.apache.sysml.hops.ParameterizedBuiltinOp;\nimport org.apache.sysml.hops.TernaryOp;\nimport org.apache.sysml.hops.UnaryOp;\n-import org.apache.sysml.hops.codegen.SpoofCompiler;\nimport org.apache.sysml.hops.codegen.cplan.CNode;\nimport org.apache.sysml.hops.codegen.cplan.CNodeBinary;\nimport org.apache.sysml.hops.codegen.cplan.CNodeBinary.BinType;\n@@ -86,8 +85,7 @@ public class TemplateRow extends TemplateBase\n&& hop.getInput().get(0).getDim1()>1 && hop.getInput().get(0).getDim2()>1)\n|| (hop instanceof AggBinaryOp && hop.dimsKnown() && LibMatrixMult.isSkinnyRightHandSide(\nhop.getInput().get(0).getDim1(), hop.getInput().get(0).getDim2(), //MM\n- hop.getInput().get(1).getDim1(), hop.getInput().get(1).getDim2(),\n- SpoofCompiler.PLAN_SEL_POLICY.isCostBased())\n+ hop.getInput().get(1).getDim1(), hop.getInput().get(1).getDim2(), false)\n&& hop.getInput().get(0).getDim1()>1 && hop.getInput().get(0).getDim2()>1\n&& !HopRewriteUtils.isOuterProductLikeMM(hop))\n|| (HopRewriteUtils.isTransposeOperation(hop) && hop.getParent().size()==1\n@@ -158,9 +156,8 @@ public class TemplateRow extends TemplateBase\n//check for fusable but not opening matrix multiply (vect_outer-mult)\nHop in1 = hop.getInput().get(0); //transpose\nHop in2 = hop.getInput().get(1);\n- boolean inclSizes = SpoofCompiler.PLAN_SEL_POLICY.isCostBased();\n- return LibMatrixMult.isSkinnyRightHandSide(in1.getDim2(), in1.getDim1(), hop.getDim1(), hop.getDim2(), inclSizes)\n- || LibMatrixMult.isSkinnyRightHandSide(in2.getDim1(), in2.getDim2(), hop.getDim2(), hop.getDim1(), inclSizes);\n+ return LibMatrixMult.isSkinnyRightHandSide(in1.getDim2(), in1.getDim1(), hop.getDim1(), hop.getDim2(), false)\n+ || LibMatrixMult.isSkinnyRightHandSide(in2.getDim1(), in2.getDim2(), hop.getDim2(), hop.getDim1(), false);\n}\nprivate static boolean isPartOfValidCumAggChain(Hop hop) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeTernary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeTernary.java", "diff": "@@ -27,7 +27,7 @@ public class CNodeTernary extends CNode\n{\npublic enum TernaryType {\nPLUS_MULT, MINUS_MULT,\n- REPLACE, REPLACE_NAN,\n+ REPLACE, REPLACE_NAN, IFELSE,\nLOOKUP_RC1, LOOKUP_RVECT1;\n@@ -53,6 +53,9 @@ public class CNodeTernary extends CNode\ncase REPLACE_NAN:\nreturn \" double %TMP% = Double.isNaN(%IN1%) ? %IN3% : %IN1%;\\n\";\n+ case IFELSE:\n+ return \" double %TMP% = (%IN1% != 0) ? %IN2% : %IN3%;\\n\";\n+\ncase LOOKUP_RC1:\nreturn sparse ?\n\" double %TMP% = getValue(%IN1v%, %IN1i%, ai, alen, %IN3%-1);\\n\" :\n@@ -128,11 +131,10 @@ public class CNodeTernary extends CNode\ncase MINUS_MULT: return \"t(-*)\";\ncase REPLACE:\ncase REPLACE_NAN: return \"t(rplc)\";\n+ case IFELSE: return \"t(ifelse)\";\ncase LOOKUP_RC1: return \"u(ixrc1)\";\ncase LOOKUP_RVECT1: return \"u(ixrv1)\";\n-\n- default:\n- return super.toString();\n+ default: return super.toString();\n}\n}\n@@ -143,6 +145,7 @@ public class CNodeTernary extends CNode\ncase MINUS_MULT:\ncase REPLACE:\ncase REPLACE_NAN:\n+ case IFELSE:\ncase LOOKUP_RC1:\n_rows = 0;\n_cols = 0;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "diff": "@@ -34,6 +34,7 @@ import org.apache.sysml.hops.Hop;\nimport org.apache.sysml.hops.UnaryOp;\nimport org.apache.sysml.hops.Hop.AggOp;\nimport org.apache.sysml.hops.Hop.OpOp2;\n+import org.apache.sysml.hops.Hop.OpOp3;\nimport org.apache.sysml.hops.Hop.ParamBuiltinOp;\nimport org.apache.sysml.hops.IndexingOp;\nimport org.apache.sysml.hops.LiteralOp;\n@@ -208,6 +209,7 @@ public class TemplateCell extends TemplateBase\n//add lookups if required\ncdata1 = TemplateUtils.wrapLookupIfNecessary(cdata1, hop.getInput().get(0));\n+ cdata2 = TemplateUtils.wrapLookupIfNecessary(cdata2, hop.getInput().get(1));\ncdata3 = TemplateUtils.wrapLookupIfNecessary(cdata3, hop.getInput().get(2));\n//construct ternary cnode, primitive operation derived from OpOp3\n@@ -299,11 +301,11 @@ public class TemplateCell extends TemplateBase\n//prepare indicators for ternary operations\nboolean isTernaryVectorScalarVector = false;\nboolean isTernaryMatrixScalarMatrixDense = false;\n+ boolean isTernaryIfElse = (HopRewriteUtils.isTernary(hop, OpOp3.IFELSE) && hop.getDataType().isMatrix());\nif( hop instanceof TernaryOp && hop.getInput().size()==3 && hop.dimsKnown()\n&& HopRewriteUtils.checkInputDataTypes(hop, DataType.MATRIX, DataType.SCALAR, DataType.MATRIX) ) {\nHop left = hop.getInput().get(0);\nHop right = hop.getInput().get(2);\n-\nisTernaryVectorScalarVector = TemplateUtils.isVector(left) && TemplateUtils.isVector(right);\nisTernaryMatrixScalarMatrixDense = HopRewriteUtils.isEqualSize(left, right)\n&& !HopRewriteUtils.isSparse(left) && !HopRewriteUtils.isSparse(right);\n@@ -312,7 +314,7 @@ public class TemplateCell extends TemplateBase\n//check supported unary, binary, ternary operations\nreturn hop.getDataType() == DataType.MATRIX && TemplateUtils.isOperationSupported(hop) && (hop instanceof UnaryOp\n|| isBinaryMatrixScalar || isBinaryMatrixVector || isBinaryMatrixMatrix\n- || isTernaryVectorScalarVector || isTernaryMatrixScalarMatrixDense\n+ || isTernaryVectorScalarVector || isTernaryMatrixScalarMatrixDense || isTernaryIfElse\n|| (hop instanceof ParameterizedBuiltinOp && ((ParameterizedBuiltinOp)hop).getOp()==ParamBuiltinOp.REPLACE));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "diff": "@@ -53,6 +53,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME15 = TEST_NAME+15; //colMins(2*log(X))\nprivate static final String TEST_NAME16 = TEST_NAME+16; //colSums(2*log(X));\nprivate static final String TEST_NAME17 = TEST_NAME+17; //xor operation\n+ private static final String TEST_NAME18 = TEST_NAME+18; //sum(ifelse(X,Y,Z))\nprivate static final String TEST_DIR = \"functions/codegen/\";\n@@ -66,7 +67,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for( int i=1; i<=17; i++ ) {\n+ for( int i=1; i<=18; i++ ) {\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(\nTEST_CLASS_DIR, TEST_NAME+i, new String[] {String.valueOf(i)}) );\n}\n@@ -304,6 +305,21 @@ public class CellwiseTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME17, true, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenCellwiseRewrite18() {\n+ testCodegenIntegration( TEST_NAME18, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwise18() {\n+ testCodegenIntegration( TEST_NAME18, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwiseRewrite18_sp() {\n+ testCodegenIntegration( TEST_NAME18, true, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl18.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+\n+X = matrix(seq(-1000, 198999), 1000, 200, byrow=TRUE);\n+Y = matrix(seq(0, 199999), 1000, 200, byrow=TRUE);\n+Z = matrix(seq(1000, 200999), 1000, 200, byrow=TRUE);\n+\n+R = as.matrix(sum(as.numeric(ifelse(X,Y,Z))));\n+\n+writeMM(as(R,\"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl18.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(-1000, 198999), 1000, 200);\n+Y = matrix(seq(0, 199999), 1000, 200);\n+Z = matrix(seq(1000, 200999), 1000, 200);\n+\n+while(FALSE){}\n+\n+R = as.matrix(sum(ifelse(X,Y,Z)));\n+\n+write(R, $1)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/mllearn/keras2caffe.py", "new_path": "src/main/python/systemml/mllearn/keras2caffe.py", "diff": "@@ -76,10 +76,8 @@ def _getInboundLayers(layer):\nfor node in inbound_nodes:\nnode_list = node.inbound_layers # get layers pointing to this node\nin_names = in_names + node_list\n- if any('flat' in s.name for s in in_names): # For Caffe2DML to reroute any use of Flatten layers\n- return _getInboundLayers([s for s in in_names if 'flat' in s.name][0])\n- return in_names\n-\n+ # For Caffe2DML to reroute any use of Flatten layers\n+ return list(chain.from_iterable( [ _getInboundLayers(l) if isinstance(l, keras.layers.Flatten) else [ l ] for l in in_names ] ))\ndef _getCompensatedAxis(layer):\ncompensated_axis = layer.axis\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/dl/CaffeLayer.scala", "new_path": "src/main/scala/org/apache/sysml/api/dl/CaffeLayer.scala", "diff": "@@ -439,6 +439,8 @@ class Concat(val param: LayerParameter, val id: Int, val net: CaffeNetwork) exte\n// This is useful because we do not support multi-input cbind and rbind in DML.\ndef _getMultiFn(fn: String): String = {\nif (_childLayers == null) _childLayers = net.getBottomLayers(param.getName).map(l => net.getCaffeLayer(l)).toList\n+ if(_childLayers.length < 2)\n+ throw new DMLRuntimeException(\"Incorrect usage of Concat layer. Expected atleast 2 bottom layers, but found \" + _childLayers.length)\nvar tmp = fn + \"(\" + _childLayers(0).out + \", \" + _childLayers(1).out + \")\"\nfor (i <- 2 until _childLayers.size) {\ntmp = fn + \"(\" + tmp + \", \" + _childLayers(i).out + \")\"\n@@ -492,20 +494,20 @@ class Concat(val param: LayerParameter, val id: Int, val net: CaffeNetwork) exte\nelse \" = \" + dOutVar + \"[,\" + indexString + \" ]; \"\n// concat_start_index = concat_end_index + 1\n- // concat_end_index = concat_start_index + $$ - 1\n+ // concat_end_index = concat_start_index + ## - 1\nval initializeIndexString = \"concat_start_index\" + outSuffix + \" = concat_end_index\" + outSuffix + \" + 1; concat_end_index\" + outSuffix +\n- \" = concat_start_index\" + outSuffix + \" + $$ - 1; \"\n+ \" = concat_start_index\" + outSuffix + \" + ## - 1; \"\nif (param.getConcatParam.getAxis == 0) {\nbottomLayers.map(l => {\ndmlScript\n- .append(initializeIndexString.replaceAll(\"$$\", nrow(l.out)))\n+ .append(initializeIndexString.replaceAll(\"##\", nrow(l.out)))\n// X1 = Z[concat_start_index:concat_end_index,]\n.append(dX(l.id) + outSuffix + doutVarAssignment)\n})\n} else {\nbottomLayers.map(l => {\ndmlScript\n- .append(initializeIndexString.replaceAll(\"$$\", int_mult(l.outputShape._1, l.outputShape._2, l.outputShape._3)))\n+ .append(initializeIndexString.replaceAll(\"##\", int_mult(l.outputShape._1, l.outputShape._2, l.outputShape._3)))\n// X1 = Z[concat_start_index:concat_end_index,]\n.append(dX(l.id) + outSuffix + doutVarAssignment)\n})\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeData.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeData.java", "diff": "@@ -60,6 +60,8 @@ public class CNodeData extends CNode\nreturn \"Double.POSITIVE_INFINITY\";\nelse if( \"-Infinity\".equals(_name) )\nreturn \"Double.NEGATIVE_INFINITY\";\n+ else if( \"true\".equals(_name) || \"false\".equals(_name) )\n+ return \"true\".equals(_name) ? \"1d\" : \"0d\";\nelse\nreturn _name;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "diff": "@@ -317,7 +317,9 @@ public class TemplateUtils\n&& !TemplateUtils.isUnary(output,\nUnaryType.EXP, UnaryType.LOG, UnaryType.ROW_COUNTNNZS))\n|| (output instanceof CNodeBinary\n- && !TemplateUtils.isBinary(output, BinType.VECT_OUTERMULT_ADD)))\n+ && !TemplateUtils.isBinary(output, BinType.VECT_OUTERMULT_ADD))\n+ || output instanceof CNodeTernary\n+ && ((CNodeTernary)output).getType() == TernaryType.IFELSE)\n&& hasOnlyDataNodeOrLookupInputs(output);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "diff": "@@ -54,7 +54,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME16 = TEST_NAME+16; //colSums(2*log(X));\nprivate static final String TEST_NAME17 = TEST_NAME+17; //xor operation\nprivate static final String TEST_NAME18 = TEST_NAME+18; //sum(ifelse(X,Y,Z))\n-\n+ private static final String TEST_NAME19 = TEST_NAME+19; //sum(ifelse(true,Y,Z))+sum(ifelse(false,Y,Z))\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + CellwiseTmplTest.class.getSimpleName() + \"/\";\n@@ -67,7 +67,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for( int i=1; i<=18; i++ ) {\n+ for( int i=1; i<=19; i++ ) {\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(\nTEST_CLASS_DIR, TEST_NAME+i, new String[] {String.valueOf(i)}) );\n}\n@@ -320,6 +320,21 @@ public class CellwiseTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME18, true, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenCellwiseRewrite19() {\n+ testCodegenIntegration( TEST_NAME19, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwise19() {\n+ testCodegenIntegration( TEST_NAME19, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwiseRewrite19_sp() {\n+ testCodegenIntegration( TEST_NAME19, true, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\n@@ -371,7 +386,8 @@ public class CellwiseTmplTest extends AutomatedTestBase\nTestUtils.compareMatrices(dmlfile, rfile, eps, \"Stat-DML\", \"Stat-R\");\n}\n- if( !(rewrites && testname.equals(TEST_NAME2)) ) //sigmoid\n+ if( !(rewrites && (testname.equals(TEST_NAME2)\n+ || testname.equals(TEST_NAME19))) ) //sigmoid\nAssert.assertTrue(heavyHittersContainsSubString(\n\"spoofCell\", \"sp_spoofCell\", \"spoofMA\", \"sp_spoofMA\"));\nif( testname.equals(TEST_NAME7) ) //ensure matrix mult is fused\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl19.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+\n+Y = as.numeric(matrix(seq(0, 199999), 1000, 200, byrow=TRUE));\n+Z = as.numeric(matrix(seq(1000, 200999), 1000, 200, byrow=TRUE));\n+R = as.matrix(sum(Y) + sum(Z));\n+\n+writeMM(as(R,\"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl19.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+Y = matrix(seq(0, 199999), 1000, 200);\n+Z = matrix(seq(1000, 200999), 1000, 200);\n+\n+while(FALSE){}\n+\n+R = as.matrix(sum(ifelse(TRUE,Y,Z)) + sum(ifelse(FALSE,Y,Z)));\n+write(R, $1)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MatrixReshapeSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MatrixReshapeSPInstruction.java", "diff": "@@ -82,8 +82,8 @@ public class MatrixReshapeSPInstruction extends UnarySPInstruction {\nSparkExecutionContext sec = (SparkExecutionContext)ec;\n//get parameters\n- int rows = (int)ec.getScalarInput(_opRows.getName(), _opRows.getValueType(), _opRows.isLiteral()).getLongValue(); //save cast\n- int cols = (int)ec.getScalarInput(_opCols.getName(), _opCols.getValueType(), _opCols.isLiteral()).getLongValue(); //save cast\n+ long rows = ec.getScalarInput(_opRows.getName(), _opRows.getValueType(), _opRows.isLiteral()).getLongValue(); //save cast\n+ long cols = ec.getScalarInput(_opCols.getName(), _opCols.getValueType(), _opCols.isLiteral()).getLongValue(); //save cast\nboolean byRow = ec.getScalarInput(_opByRow.getName(), ValueType.BOOLEAN, _opByRow.isLiteral()).getBooleanValue();\n//get inputs\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/CompressedMatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/CompressedMatrixBlock.java", "diff": "@@ -128,7 +128,7 @@ public class CompressedMatrixBlock extends MatrixBlock implements Externalizable\nprotected boolean _sharedDDC1Dict = false;\npublic CompressedMatrixBlock() {\n- super(-1, -1, true);\n+ super(0, 0, true);\n}\n/**\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/conf/ConfigurationManager.java", "new_path": "src/main/java/org/apache/sysml/conf/ConfigurationManager.java", "diff": "package org.apache.sysml.conf;\nimport org.apache.hadoop.mapred.JobConf;\n+import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.conf.CompilerConfig.ConfigType;\nimport org.apache.sysml.runtime.matrix.mapred.MRConfigurationNames;\nimport org.apache.sysml.runtime.matrix.mapred.MRJobConfiguration;\n@@ -190,11 +191,14 @@ public class ConfigurationManager\n}\npublic static boolean isCodegenEnabled() {\n- return getDMLConfig().getBooleanValue(DMLConfig.CODEGEN)\n- || getCompilerConfigFlag(ConfigType.CODEGEN_ENABLED);\n+ return (getDMLConfig().getBooleanValue(DMLConfig.CODEGEN)\n+ || getCompilerConfigFlag(ConfigType.CODEGEN_ENABLED))\n+ && !DMLScript.USE_ACCELERATOR;\n+ //note: until codegen is supported for the GPU backend, we globally\n+ //disable codegen if operations are forced to the GPU to avoid\n+ //a counter-productive impact on performance.\n}\n-\n///////////////////////////////////////\n// Thread-local classes\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeUnary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeUnary.java", "diff": "@@ -36,6 +36,7 @@ public class CNodeUnary extends CNode\nVECT_SIN, VECT_COS, VECT_TAN, VECT_ASIN, VECT_ACOS, VECT_ATAN,\nVECT_SINH, VECT_COSH, VECT_TANH,\nVECT_CUMSUM, VECT_CUMMIN, VECT_CUMMAX,\n+ VECT_SPROP, VECT_SIGMOID,\nEXP, POW2, MULT2, SQRT, LOG, LOG_NZ,\nABS, ROUND, CEIL, FLOOR, SIGN,\nSIN, COS, TAN, ASIN, ACOS, ATAN, SINH, COSH, TANH,\n@@ -81,7 +82,9 @@ public class CNodeUnary extends CNode\ncase VECT_TANH:\ncase VECT_CUMSUM:\ncase VECT_CUMMIN:\n- case VECT_CUMMAX:{\n+ case VECT_CUMMAX:\n+ case VECT_SPROP:\n+ case VECT_SIGMOID: {\nString vectName = getVectorPrimitiveName();\nreturn sparse ? \" double[] %TMP% = LibSpoofPrimitives.vect\"+vectName+\"Write(%IN1v%, %IN1i%, %POS1%, alen, len);\\n\" :\n\" double[] %TMP% = LibSpoofPrimitives.vect\"+vectName+\"Write(%IN1%, %POS1%, %LEN%);\\n\";\n@@ -155,8 +158,8 @@ public class CNodeUnary extends CNode\n|| this == VECT_SIN || this == VECT_COS || this == VECT_TAN\n|| this == VECT_ASIN || this == VECT_ACOS || this == VECT_ATAN\n|| this == VECT_SINH || this == VECT_COSH || this == VECT_TANH\n- || this == VECT_CUMSUM || this == VECT_CUMMIN\n- || this == VECT_CUMMAX;\n+ || this == VECT_CUMSUM || this == VECT_CUMMIN || this == VECT_CUMMAX\n+ || this == VECT_SPROP || this == VECT_SIGMOID;\n}\npublic UnaryType getVectorAddPrimitive() {\nreturn UnaryType.valueOf(\"VECT_\"+getVectorPrimitiveName().toUpperCase()+\"_ADD\");\n@@ -267,7 +270,9 @@ public class CNodeUnary extends CNode\ncase VECT_CUMSUM:\ncase VECT_CUMMIN:\ncase VECT_CUMMAX:\n- case VECT_SIGN: return \"u(v\"+_type.name().toLowerCase()+\")\";\n+ case VECT_SIGN:\n+ case VECT_SIGMOID:\n+ case VECT_SPROP:return \"u(v\"+_type.name().toLowerCase()+\")\";\ncase LOOKUP_R: return \"u(ixr)\";\ncase LOOKUP_C: return \"u(ixc)\";\ncase LOOKUP_RC: return \"u(ixrc)\";\n@@ -302,6 +307,8 @@ public class CNodeUnary extends CNode\ncase VECT_CUMSUM:\ncase VECT_CUMMIN:\ncase VECT_CUMMAX:\n+ case VECT_SPROP:\n+ case VECT_SIGMOID:\n_rows = _inputs.get(0)._rows;\n_cols = _inputs.get(0)._cols;\n_dataType= DataType.MATRIX;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -63,7 +63,7 @@ public class TemplateRow extends TemplateBase\nprivate static final Hop.OpOp1[] SUPPORTED_VECT_UNARY = new OpOp1[]{\nOpOp1.EXP, OpOp1.SQRT, OpOp1.LOG, OpOp1.ABS, OpOp1.ROUND, OpOp1.CEIL, OpOp1.FLOOR, OpOp1.SIGN,\nOpOp1.SIN, OpOp1.COS, OpOp1.TAN, OpOp1.ASIN, OpOp1.ACOS, OpOp1.ATAN, OpOp1.SINH, OpOp1.COSH, OpOp1.TANH,\n- OpOp1.CUMSUM, OpOp1.CUMMIN, OpOp1.CUMMAX};\n+ OpOp1.CUMSUM, OpOp1.CUMMIN, OpOp1.CUMMAX, OpOp1.SPROP, OpOp1.SIGMOID};\nprivate static final Hop.OpOp2[] SUPPORTED_VECT_BINARY = new OpOp2[]{\nOpOp2.MULT, OpOp2.DIV, OpOp2.MINUS, OpOp2.PLUS, OpOp2.POW, OpOp2.MIN, OpOp2.MAX, OpOp2.XOR,\nOpOp2.EQUAL, OpOp2.NOTEQUAL, OpOp2.LESS, OpOp2.LESSEQUAL, OpOp2.GREATER, OpOp2.GREATEREQUAL};\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "diff": "@@ -1505,6 +1505,58 @@ public class LibSpoofPrimitives\nreturn c;\n}\n+ //custom sprop\n+\n+ public static void vectSpropAdd(double[] a, double[] c, int ai, int ci, int len) {\n+ for( int j = ai; j < ai+len; j++, ci++)\n+ c[ci] += a[j] * (1 - a[j]);\n+ }\n+\n+ public static void vectSpropAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[ci + aix[j]] += a[j] * (1 - a[j]);\n+ }\n+\n+ public static double[] vectSpropWrite(double[] a, int ai, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++, ai++)\n+ c[j] = a[j] * (1 - a[j]);\n+ return c;\n+ }\n+\n+ public static double[] vectSpropWrite(double[] a, int[] aix, int ai, int alen, int len) {\n+ double[] c = allocVector(len, true);\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[aix[j]] = a[j] * (1 - a[j]);\n+ return c;\n+ }\n+\n+ //custom sigmoid\n+\n+ public static void vectSigmoidAdd(double[] a, double[] c, int ai, int ci, int len) {\n+ for( int j = ai; j < ai+len; j++, ci++)\n+ c[ci] += 1 / (1 + FastMath.exp(-a[j]));\n+ }\n+\n+ public static void vectSigmoidAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[ci + aix[j]] += 1 / (1 + FastMath.exp(-a[j]));\n+ }\n+\n+ public static double[] vectSigmoidWrite(double[] a, int ai, int len) {\n+ double[] c = allocVector(len, false);\n+ for( int j = 0; j < len; j++, ai++)\n+ c[j] = 1 / (1 + FastMath.exp(-a[j]));\n+ return c;\n+ }\n+\n+ public static double[] vectSigmoidWrite(double[] a, int[] aix, int ai, int alen, int len) {\n+ double[] c = allocVector(len, true, 0.5); //sigmoid(0) = 0.5\n+ for( int j = ai; j < ai+alen; j++ )\n+ c[aix[j]] = 1 / (1 + FastMath.exp(-a[j]));\n+ return c;\n+ }\n+\n//custom vector equal\npublic static void vectEqualAdd(double[] a, double bval, double[] c, int ai, int ci, int len) {\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "diff": "@@ -73,6 +73,8 @@ public class RowAggTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME34 = TEST_NAME+\"34\"; //X / rowSums(X!=0)\nprivate static final String TEST_NAME35 = TEST_NAME+\"35\"; //cbind(X/rowSums(X), Y, Z)\nprivate static final String TEST_NAME36 = TEST_NAME+\"36\"; //xor operation\n+ private static final String TEST_NAME37 = TEST_NAME+\"37\"; //sprop(X/rowSums)\n+ private static final String TEST_NAME38 = TEST_NAME+\"38\"; //sigmoid(X/rowSums)\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + RowAggTmplTest.class.getSimpleName() + \"/\";\n@@ -84,7 +86,7 @@ public class RowAggTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for(int i=1; i<=36; i++)\n+ for(int i=1; i<=38; i++)\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME+i, new String[] { String.valueOf(i) }) );\n}\n@@ -628,6 +630,36 @@ public class RowAggTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME36, false, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenRowAggRewrite37CP() {\n+ testCodegenIntegration( TEST_NAME37, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg37CP() {\n+ testCodegenIntegration( TEST_NAME37, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg37SP() {\n+ testCodegenIntegration( TEST_NAME37, false, ExecType.SPARK );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAggRewrite38CP() {\n+ testCodegenIntegration( TEST_NAME38, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg38CP() {\n+ testCodegenIntegration( TEST_NAME38, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg38SP() {\n+ testCodegenIntegration( TEST_NAME38, false, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\nboolean oldFlag = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern37.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+library(\"matrixStats\")\n+\n+X = matrix(seq(1, 6000)/600, 300, 20, byrow=TRUE);\n+\n+Y = X/(rowSums(X)%*%matrix(1,1,ncol(X)))\n+S = Y * (Y-1);\n+\n+writeMM(as(S, \"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern37.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(1, 6000)/600, 300, 20);\n+\n+Y = X/rowSums(X)\n+S = Y * (Y-1);\n+\n+write(S, $1);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern38.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+library(\"matrixStats\")\n+\n+X = matrix(seq(1, 6000)/600, 300, 20, byrow=TRUE);\n+\n+Y = X/(rowSums(X)%*%matrix(1,1,ncol(X)))\n+S = 1 / (1 + exp(-Y));\n+\n+writeMM(as(S, \"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern38.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(1, 6000)/600, 300, 20);\n+\n+Y = X/rowSums(X)\n+S = 1 / (1 + exp(-Y));\n+\n+write(S, $1);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -197,14 +197,13 @@ public class OperationsOnMatrixValues\n//set initial values according to operator\nif(op.initialValue==0) {\n- valueOut.reset(outRow, outCol, sparseHint);\n- correction.reset(corRow, corCol, false);\n+ valueOut.reset(Math.max(outRow,0), Math.max(outCol,0), sparseHint);\n+ correction.reset(Math.max(corRow,0), Math.max(corCol,0), false);\n}\nelse {\n- valueOut.reset(outRow, outCol, op.initialValue);\n- correction.reset(corRow, corCol, op.initialValue);\n+ valueOut.reset(Math.max(outRow, 0), Math.max(outCol,0), op.initialValue);\n+ correction.reset(Math.max(corRow,0), Math.max(corCol,0), op.initialValue);\n}\n-\n}\nelse\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "diff": "@@ -210,7 +210,7 @@ public class CNodeBinary extends CNode\ncase XOR:\nreturn \" double %TMP% = ( (%IN1% != 0) != (%IN2% != 0) ) ? 1 : 0;\\n\";\ncase BITWAND:\n- return \" double %TMP% = LibSpoofPrimitives.intDiv(%IN1%, %IN2%);\\n\";\n+ return \" double %TMP% = LibSpoofPrimitives.bwAnd(%IN1%, %IN2%);\\n\";\ndefault:\nthrow new RuntimeException(\"Invalid binary type: \"+this.toString());\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "diff": "@@ -139,6 +139,26 @@ public class PreparedScript implements ConfigurableAPI\n}\n}\n+ /**\n+ * Get the dml configuration object associated with\n+ * the prepared script instance.\n+ *\n+ * @return dml configuration\n+ */\n+ public DMLConfig getDMLConfig() {\n+ return _dmlconf;\n+ }\n+\n+ /**\n+ * Get the compiler configuration object associated with\n+ * the prepared script instance.\n+ *\n+ * @return compiler configuration\n+ */\n+ public CompilerConfig getCompilerConfig() {\n+ return _cconf;\n+ }\n+\n/**\n* Binds a scalar boolean to a registered input variable.\n*\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/algorithms/l2-svm.dml", "new_path": "scripts/algorithms/l2-svm.dml", "diff": "# Log String --- [OPTIONAL] Location to write the log file\n# ---------------------------------------------------------------------------------------------\n-# hadoop jar SystemML.jar -f $L2SVM_HOME/l2-svm.dml -nvargs X=$INPUT_DIR/X Y=$INPUT_DIR/Y icpt=0 tol=0.001 reg=1 maxiter=100 model=$OUPUT_DIR/w Log=$OUTPUT_DIR/Log fmt=\"text\"\n+# hadoop jar SystemML.jar -f $L2SVM_HOME/l2-svm.dml -nvargs X=$INPUT_DIR/X Y=$INPUT_DIR/Y \\\n+# icpt=0 tol=0.001 reg=1 maxiter=100 model=$OUPUT_DIR/w Log=$OUTPUT_DIR/Log fmt=\"text\"\n#\n# Note about inputs:\n-# Assumes that labels (entries in Y)\n-# are set to either -1 or +1\n-# or the result of recoding\n-#\n+# Assumes that labels (entries in Y) are set to either -1 or +1 or non-negative integers\n-cmdLine_fmt = ifdef($fmt, \"text\")\n-cmdLine_icpt = ifdef($icpt, 0)\n-cmdLine_tol = ifdef($tol, 0.001)\n-cmdLine_reg = ifdef($reg, 1.0)\n-cmdLine_maxiter = ifdef($maxiter, 100)\n+fmt = ifdef($fmt, \"text\")\n+intercept = ifdef($icpt, 0)\n+epsilon = ifdef($tol, 0.001)\n+lambda = ifdef($reg, 1.0)\n+maxiterations = ifdef($maxiter, 100)\nX = read($X)\nY = read($Y)\n+#check input parameter assertions\nif(nrow(X) < 2)\nstop(\"Stopping due to invalid inputs: Not possible to learn a binary class classifier without at least 2 rows\")\n+if(intercept != 0 & intercept != 1)\n+ stop(\"Stopping due to invalid argument: Currently supported intercept options are 0 and 1\")\n+if(epsilon < 0)\n+ stop(\"Stopping due to invalid argument: Tolerance (tol) must be non-negative\")\n+if(lambda < 0)\n+ stop(\"Stopping due to invalid argument: Regularization constant (reg) must be non-negative\")\n+if(maxiterations < 1)\n+ stop(\"Stopping due to invalid argument: Maximum iterations should be a positive integer\")\n+#check input lables and transform into -1/1\ncheck_min = min(Y)\ncheck_max = max(Y)\nnum_min = sum(Y == check_min)\nnum_max = sum(Y == check_max)\n-\nif(check_min == check_max)\nstop(\"Stopping due to invalid inputs: Y seems to contain exactly one label\")\n-\nif(num_min + num_max != nrow(Y))\nstop(\"Stopping due to invalid inputs: Y seems to contain more than 2 labels\")\n-\n-if(check_min != -1 | check_max != +1)\n+if(check_min != -1 | check_max != 1)\nY = 2/(check_max - check_min)*Y - (check_min + check_max)/(check_max - check_min)\npositive_label = check_max\nnegative_label = check_min\n-\n-intercept = cmdLine_icpt\n-if(intercept != 0 & intercept != 1)\n- stop(\"Stopping due to invalid argument: Currently supported intercept options are 0 and 1\")\n-\n-epsilon = cmdLine_tol\n-if(epsilon < 0)\n- stop(\"Stopping due to invalid argument: Tolerance (tol) must be non-negative\")\n-\n-lambda = cmdLine_reg\n-if(lambda < 0)\n- stop(\"Stopping due to invalid argument: Regularization constant (reg) must be non-negative\")\n-\n-maxiterations = cmdLine_maxiter\n-if(maxiterations < 1)\n- stop(\"Stopping due to invalid argument: Maximum iterations should be a positive integer\")\n-\nnum_samples = nrow(X)\ndimensions = ncol(X)\n+num_rows_in_w = dimensions\nif (intercept == 1) {\nones = matrix(1, rows=num_samples, cols=1)\nX = cbind(X, ones);\n+ num_rows_in_w += 1\n}\n-num_rows_in_w = dimensions\n-if(intercept == 1){\n- num_rows_in_w = num_rows_in_w + 1\n-}\n-w = matrix(0, rows=num_rows_in_w, cols=1)\n-\n+w = matrix(0, num_rows_in_w, 1)\n+Xw = matrix(0, rows=nrow(X), cols=1)\ng_old = t(X) %*% Y\ns = g_old\n-Xw = matrix(0, rows=nrow(X), cols=1)\ndebug_str = \"# Iter, Obj\"\niter = 0\ncontinue = TRUE\n+\nwhile(continue & iter < maxiterations) {\n# minimizing primal obj along direction s\nstep_sz = 0\n@@ -127,13 +113,12 @@ while(continue & iter < maxiterations) {\ncontinue1 = TRUE\nwhile(continue1) {\ntmp_Xw = Xw + step_sz*Xd\n- out = 1 - Y * (tmp_Xw)\n- sv = (out > 0)\n+ out = 1 - Y * tmp_Xw\n+ sv = out > 0\nout = out * sv\ng = wd + step_sz*dd - sum(out * Y * Xd)\nh = dd + sum(Xd * sv * Xd)\nstep_sz = step_sz - g/h\n-\ncontinue1 = (g*g/h >= 0.0000000001);\n}\n@@ -142,7 +127,7 @@ while(continue & iter < maxiterations) {\nXw += step_sz * Xd\nout = 1 - Y * Xw\n- sv = (out > 0)\n+ sv = out > 0\nout = sv * out\nobj = 0.5 * sum(out * out) + lambda/2 * sum(w * w)\ng_new = t(X) %*% (out * Y) - lambda * w\n@@ -161,16 +146,15 @@ while(continue & iter < maxiterations) {\niter = iter + 1\n}\n-extra_model_params = matrix(0, rows=4, cols=1)\n+extra_model_params = matrix(0, 4, 1)\nextra_model_params[1,1] = positive_label\nextra_model_params[2,1] = negative_label\nextra_model_params[3,1] = intercept\nextra_model_params[4,1] = dimensions\n-w = t(cbind(t(w), t(extra_model_params)))\n-write(w, $model, format=cmdLine_fmt)\n+w = rbind(w, extra_model_params)\n+write(w, $model, format=fmt)\nlogFile = $Log\n-if(logFile != \" \") {\n+if(logFile != \" \")\nwrite(debug_str, logFile)\n-}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "diff": "@@ -147,9 +147,13 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\nmemo.setDistinct(e.getKey(), e.getValue());\n//maintain statistics\n- if( DMLScript.STATISTICS )\n+ if( DMLScript.STATISTICS ) {\n+ if( sumMatPoints >= 63 )\n+ LOG.warn(\"Long overflow on maintaining codegen statistics \"\n+ + \"for a DAG with \"+sumMatPoints+\" interesting points.\");\nStatistics.incrementCodegenEnumAll(UtilFunctions.pow(2, sumMatPoints));\n}\n+ }\nprivate void selectPlans(CPlanMemoTable memo, PlanPartition part)\n{\n@@ -1033,6 +1037,7 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\n}\nelse if( current instanceof TernaryOp ) {\nswitch( ((TernaryOp)current).getOp() ) {\n+ case IFELSE:\ncase PLUS_MULT:\ncase MINUS_MULT: costs = 2; break;\ncase CTABLE: costs = 3; break;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "diff": "@@ -63,12 +63,15 @@ public class CNodeBinary extends CNode\npublic boolean isCommutative() {\nboolean ssComm = (this==EQUAL || this==NOTEQUAL\n- || this==PLUS || this==MULT || this==MIN || this==MAX);\n+ || this==PLUS || this==MULT || this==MIN || this==MAX\n+ || this==OR || this==AND || this==XOR || this==BITWAND);\nboolean vsComm = (this==VECT_EQUAL_SCALAR || this==VECT_NOTEQUAL_SCALAR\n|| this==VECT_PLUS_SCALAR || this==VECT_MULT_SCALAR\n- || this==VECT_MIN_SCALAR || this==VECT_MAX_SCALAR);\n+ || this==VECT_MIN_SCALAR || this==VECT_MAX_SCALAR\n+ || this==VECT_XOR_SCALAR || this==VECT_BITWAND_SCALAR );\nboolean vvComm = (this==VECT_EQUAL || this==VECT_NOTEQUAL\n- || this==VECT_PLUS || this==VECT_MULT || this==VECT_MIN || this==VECT_MAX);\n+ || this==VECT_PLUS || this==VECT_MULT || this==VECT_MIN || this==VECT_MAX\n+ || this==VECT_XOR || this==BinType.VECT_BITWAND);\nreturn ssComm || vsComm || vvComm;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "diff": "@@ -580,8 +580,7 @@ public class LibSpoofPrimitives\n}\npublic static void vectXorAdd(double bval, double[] a, double[] c, int ai, int ci, int len) {\n- for( int j = ai; j < ai+len; j++, ci++)\n- c[ci] += ( (bval != 0) != (a[j] != 0) ) ? 1 : 0;\n+ vectXorAdd(a, bval, c, ai, ci, len);\n}\npublic static void vectXorAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n@@ -590,8 +589,7 @@ public class LibSpoofPrimitives\n}\npublic static void vectXorAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n- for( int j = ai; j < ai+alen; j++ )\n- c[ci + aix[j]] += ( (bval != 0) != (a[j] != 0) ) ? 1 : 0;\n+ vectXorAdd(a, bval, c, aix, ai, ci, alen, len);\n}\n//1. scalar vs. dense vector\n@@ -604,10 +602,7 @@ public class LibSpoofPrimitives\n//2. dense vector vs. scalar\npublic static double[] vectXorWrite(double bval, double[] a, int ai, int len) {\n- double[] c = allocVector(len, false);\n- for( int j = 0; j < len; j++)\n- c[j] = ( (bval != 0) != (a[ai + j] != 0) ) ? 1 : 0;\n- return c;\n+ return vectXorWrite(a, bval, ai, len);\n}\n//3. dense vector vs. dense vector\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/_config.yml", "new_path": "docs/_config.yml", "diff": "@@ -15,7 +15,7 @@ exclude:\n- lang-ref\n# These allow the documentation to be updated with newer releases\n-SYSTEMML_VERSION: 1.0.0-SNAPSHOT\n+SYSTEMML_VERSION: 1.1.0-SNAPSHOT\n# if 'analytics_on' is true, analytics section will be rendered on the HTML pages\nanalytics_on: true\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/CellwiseTmplTest.java", "diff": "@@ -56,6 +56,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME18 = TEST_NAME+18; //sum(ifelse(X,Y,Z))\nprivate static final String TEST_NAME19 = TEST_NAME+19; //sum(ifelse(true,Y,Z))+sum(ifelse(false,Y,Z))\nprivate static final String TEST_NAME20 = TEST_NAME+20; //bitwAnd() operation\n+ private static final String TEST_NAME21 = TEST_NAME+21; //relu operation, (X>0)*dout\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + CellwiseTmplTest.class.getSimpleName() + \"/\";\n@@ -68,7 +69,7 @@ public class CellwiseTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for( int i=1; i<=20; i++ ) {\n+ for( int i=1; i<=21; i++ ) {\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(\nTEST_CLASS_DIR, TEST_NAME+i, new String[] {String.valueOf(i)}) );\n}\n@@ -351,6 +352,21 @@ public class CellwiseTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME20, true, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenCellwiseRewrite21() {\n+ testCodegenIntegration( TEST_NAME21, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwise21() {\n+ testCodegenIntegration( TEST_NAME21, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenCellwiseRewrite21_sp() {\n+ testCodegenIntegration( TEST_NAME21, true, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\nboolean oldRewrites = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/RowAggTmplTest.java", "diff": "@@ -76,6 +76,7 @@ public class RowAggTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME37 = TEST_NAME+\"37\"; //sprop(X/rowSums)\nprivate static final String TEST_NAME38 = TEST_NAME+\"38\"; //sigmoid(X/rowSums)\nprivate static final String TEST_NAME39 = TEST_NAME+\"39\"; //BitwAnd operation\n+ private static final String TEST_NAME40 = TEST_NAME+\"40\"; //relu operation -> (X>0)* dout\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + RowAggTmplTest.class.getSimpleName() + \"/\";\n@@ -87,7 +88,7 @@ public class RowAggTmplTest extends AutomatedTestBase\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\n- for(int i=1; i<=39; i++)\n+ for(int i=1; i<=40; i++)\naddTestConfiguration( TEST_NAME+i, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME+i, new String[] { String.valueOf(i) }) );\n}\n@@ -676,6 +677,21 @@ public class RowAggTmplTest extends AutomatedTestBase\ntestCodegenIntegration( TEST_NAME39, false, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenRowAggRewrite40CP() {\n+ testCodegenIntegration( TEST_NAME40, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg40CP() {\n+ testCodegenIntegration( TEST_NAME40, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenRowAgg40SP() {\n+ testCodegenIntegration( TEST_NAME40, false, ExecType.SPARK );\n+ }\n+\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\nboolean oldFlag = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl21.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+\n+X = matrix(seq(7, 1006), 500, 2, byrow=TRUE);\n+\n+R1 = (X/3) %% 0.6;\n+R2 = (X/3) %/% 0.6;\n+R = (R1 > 0) * R2\n+\n+writeMM(as(R,\"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/cellwisetmpl21.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(7, 1006), 500, 2);\n+\n+R1 = (X/3) %% 0.6;\n+R2 = (X/3) %/% 0.6;\n+R = (R1 > 0) * R2;\n+\n+write(R, $1)\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern40.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+library(\"matrixStats\")\n+\n+X = matrix(seq(1, 6000)/600, 300, 20, byrow=TRUE);\n+\n+Y = X/(rowSums(X)%*%matrix(1,1,ncol(X)))\n+\n+R = (X>0) * Y;\n+\n+writeMM(as(R, \"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/rowAggPattern40.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(seq(1, 6000)/600, 300, 20);\n+\n+Y = X/rowSums(X)\n+\n+R = (X>0) * Y;\n+\n+write(R, $1)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "diff": "@@ -1082,10 +1082,13 @@ public abstract class SpoofCellwise extends SpoofOperator implements Serializabl\nif( sparseSafe && b.length==0 && !a.hasUncompressedColGroup() ) {\n//note: all remaining groups are guaranteed ColGroupValue\nboolean entireGrp = (rl==0 && ru==a.getNumRows());\n+ int maxNumVals = a.getColGroups().stream().mapToInt(\n+ g -> ((ColGroupValue)g).getNumValues()).max().orElse(0);\n+ int[] counts = new int[maxNumVals];\nfor( ColGroup grp : a.getColGroups() ) {\nColGroupValue grpv = (ColGroupValue) grp;\n- int[] counts = entireGrp ?\n- grpv.getCounts() : grpv.getCounts(rl, ru);\n+ counts = entireGrp ? grpv.getCounts(counts) :\n+ grpv.getCounts(rl, ru, counts);\nfor(int k=0; k<grpv.getNumValues(); k++) {\nkbuff2.set(0, 0);\ndouble in = grpv.sumValues(k, kplus, kbuff2);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupDDC1.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupDDC1.java", "diff": "@@ -204,14 +204,14 @@ public class ColGroupDDC1 extends ColGroupDDC\n}\n@Override\n- public int[] getCounts() {\n- return getCounts(0, getNumRows());\n+ public int[] getCounts(int[] counts) {\n+ return getCounts(0, getNumRows(), counts);\n}\n@Override\n- public int[] getCounts(int rl, int ru) {\n+ public int[] getCounts(int rl, int ru, int[] counts) {\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor( int i=rl; i<ru; i++ )\ncounts[_data[i]&0xFF] ++;\nreturn counts;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupDDC2.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupDDC2.java", "diff": "@@ -193,14 +193,14 @@ public class ColGroupDDC2 extends ColGroupDDC\n}\n@Override\n- public int[] getCounts() {\n- return getCounts(0, getNumRows());\n+ public int[] getCounts(int[] counts) {\n+ return getCounts(0, getNumRows(), counts);\n}\n@Override\n- public int[] getCounts(int rl, int ru) {\n+ public int[] getCounts(int rl, int ru, int[] counts) {\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor( int i=rl; i<ru; i++ )\ncounts[_data[i]] ++;\nreturn counts;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupOLE.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupOLE.java", "diff": "@@ -229,9 +229,9 @@ public class ColGroupOLE extends ColGroupOffset\n}\n@Override\n- public int[] getCounts() {\n+ public int[] getCounts(int[] counts) {\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor (int k = 0; k < numVals; k++) {\nint boff = _ptr[k];\nint blen = len(k);\n@@ -245,10 +245,10 @@ public class ColGroupOLE extends ColGroupOffset\n}\n@Override\n- public int[] getCounts(int rl, int ru) {\n+ public int[] getCounts(int rl, int ru, int[] counts) {\nfinal int blksz = BitmapEncoder.BITMAP_BLOCK_SZ;\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor (int k = 0; k < numVals; k++) {\nint boff = _ptr[k];\nint blen = len(k);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupRLE.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupRLE.java", "diff": "@@ -225,9 +225,9 @@ public class ColGroupRLE extends ColGroupOffset\n}\n@Override\n- public int[] getCounts() {\n+ public int[] getCounts(int[] counts) {\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor (int k = 0; k < numVals; k++) {\nint boff = _ptr[k];\nint blen = len(k);\n@@ -244,9 +244,9 @@ public class ColGroupRLE extends ColGroupOffset\n}\n@Override\n- public int[] getCounts(int rl, int ru) {\n+ public int[] getCounts(int rl, int ru, int[] counts) {\nfinal int numVals = getNumValues();\n- int[] counts = new int[numVals];\n+ Arrays.fill(counts, 0, numVals, 0);\nfor (int k = 0; k < numVals; k++) {\nint boff = _ptr[k];\nint blen = len(k);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupValue.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/ColGroupValue.java", "diff": "@@ -149,9 +149,19 @@ public abstract class ColGroupValue extends ColGroup\nreturn ret;\n}\n- public abstract int[] getCounts();\n+ public final int[] getCounts() {\n+ int[] tmp = new int[getNumValues()];\n+ return getCounts(tmp);\n+ }\n+\n+ public abstract int[] getCounts(int[] out);\n+\n+ public final int[] getCounts(int rl, int ru) {\n+ int[] tmp = new int[getNumValues()];\n+ return getCounts(rl, ru, tmp);\n+ }\n- public abstract int[] getCounts(int rl, int ru);\n+ public abstract int[] getCounts(int rl, int ru, int[] out);\npublic int[] getCounts(boolean inclZeros) {\nint[] counts = getCounts();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "diff": "@@ -52,13 +52,6 @@ public class SparseBlockCSR extends SparseBlock\nprivate double[] _values = null; //value array (size: >=nnz)\nprivate int _size = 0; //actual number of nnz\n- //matrix meta data\n- protected int rlen = -1;\n- protected int clen = -1;\n- protected boolean sparse = true;\n- protected long nonZeros = 0;\n-\n-\npublic SparseBlockCSR(int rlen) {\nthis(rlen, INIT_CAPACITY);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/compress/CompressedMatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/compress/CompressedMatrixBlock.java", "diff": "@@ -300,6 +300,15 @@ public class CompressedMatrixBlock extends MatrixBlock implements Externalizable\n}\n}\n+ if( LOG.isTraceEnabled() ) {\n+ LOG.trace(\"C: \"+Arrays.toString(colsC.toArray(new Integer[0])));\n+ LOG.trace(\"-- compression ratios: \"+Arrays.toString(\n+ colsC.stream().map(c -> compRatios.get(c)).toArray()));\n+ LOG.trace(\"UC: \"+Arrays.toString(colsUC.toArray(new Integer[0])));\n+ LOG.trace(\"-- compression ratios: \"+Arrays.toString(\n+ colsUC.stream().map(c -> compRatios.get(c)).toArray()));\n+ }\n+\nif( LOG.isDebugEnabled() ) {\n_stats.timePhase1 = time.stop();\nLOG.debug(\"Compression statistics:\");\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "diff": "@@ -175,6 +175,7 @@ public class RewriteAlgebraicSimplificationStatic extends HopRewriteRule\nhi = fuseOrderOperationChain(hi); //e.g., order(order(X,2),1) -> order(X,(12))\nhi = removeUnnecessaryReorgOperation(hop, hi, i); //e.g., t(t(X))->X; rev(rev(X))->X potentially introduced by other rewrites\nhi = simplifyTransposeAggBinBinaryChains(hop, hi, i);//e.g., t(t(A)%*%t(B)+C) -> B%*%A+t(C)\n+ hi = simplifyReplaceZeroOperation(hop, hi, i); //e.g., X + (X==0) * s -> replace(X, 0, s)\nhi = removeUnnecessaryMinus(hop, hi, i); //e.g., -(-X)->X; potentially introduced by simplify binary or dyn rewrites\nhi = simplifyGroupedAggregate(hi); //e.g., aggregate(target=X,groups=y,fn=\"count\") -> aggregate(target=y,groups=y,fn=\"count\")\nif(OptimizerUtils.ALLOW_OPERATOR_FUSION) {\n@@ -1585,6 +1586,29 @@ public class RewriteAlgebraicSimplificationStatic extends HopRewriteRule\nreturn hi;\n}\n+ // Patterns: X + (X==0) * s -> replace(X, 0, s)\n+ private static Hop simplifyReplaceZeroOperation(Hop parent, Hop hi, int pos)\n+ throws HopsException\n+ {\n+ if( HopRewriteUtils.isBinary(hi, OpOp2.PLUS) && hi.getInput().get(0).isMatrix()\n+ && HopRewriteUtils.isBinary(hi.getInput().get(1), OpOp2.MULT)\n+ && hi.getInput().get(1).getInput().get(1).isScalar()\n+ && HopRewriteUtils.isBinaryMatrixScalar(hi.getInput().get(1).getInput().get(0), OpOp2.EQUAL, 0)\n+ && hi.getInput().get(1).getInput().get(0).getInput().contains(hi.getInput().get(0)) )\n+ {\n+ HashMap<String, Hop> args = new HashMap<>();\n+ args.put(\"target\", hi.getInput().get(0));\n+ args.put(\"pattern\", new LiteralOp(0));\n+ args.put(\"replacement\", hi.getInput().get(1).getInput().get(1));\n+ Hop replace = HopRewriteUtils.createParameterizedBuiltinOp(\n+ hi.getInput().get(0), args, ParamBuiltinOp.REPLACE);\n+ HopRewriteUtils.replaceChildReference(parent, hi, replace, pos);\n+ hi = replace;\n+ LOG.debug(\"Applied simplifyReplaceZeroOperation (line \"+hi.getBeginLine()+\").\");\n+ }\n+ return hi;\n+ }\n+\n/**\n* Pattners: t(t(X)) -> X, rev(rev(X)) -> X\n*\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/LibSpoofPrimitives.java", "diff": "@@ -655,15 +655,16 @@ public class LibSpoofPrimitives\n}\npublic static void vectPowAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n- if( bval == 0 ) //handle 0^0=1\n+ if( bval == 0 ) //handle 0^0=1 & a^0=1\nfor( int j=0; j<len; j++ )\nc[ci + j] += 1;\n+ else //handle 0^b=0 & a^b\nfor( int j = ai; j < ai+alen; j++ )\n- c[ci + aix[j]] += Math.pow(a[j], bval) - 1;\n+ c[ci + aix[j]] += Math.pow(a[j], bval);\n}\npublic static void vectPowAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) {\n- for( int j=0; j<len; j++ )\n+ for( int j=0; j<len; j++ ) //handle 0^0=1 & b^0=1\nc[ci + j] += 1;\nfor( int j = ai; j < ai+alen; j++ )\nc[ci + aix[j]] += Math.pow(bval, a[j]) - 1;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/ParForStatementBlock.java", "new_path": "src/main/java/org/apache/sysml/parser/ParForStatementBlock.java", "diff": "@@ -1956,9 +1956,12 @@ public class ParForStatementBlock extends ForStatementBlock\nboolean ret = ( _b.length == f2._b.length );\nfor( int i=0; i<_b.length && ret; i++ ) {\nret &= (_b[i] == f2._b[i] );\n- ret &= (_vars[i].equals(f2._vars[i])\n- ||(_vars[i].startsWith(INTERAL_FN_INDEX_ROW) && f2._vars[i].startsWith(INTERAL_FN_INDEX_ROW))\n- ||(_vars[i].startsWith(INTERAL_FN_INDEX_COL) && f2._vars[i].startsWith(INTERAL_FN_INDEX_COL)) ) ;\n+ //note robustness for null var names\n+ String var1 = String.valueOf(_vars[i]);\n+ String var2 = String.valueOf(f2._vars[i]);\n+ ret &= (var1.equals(var2)\n+ ||(var1.startsWith(INTERAL_FN_INDEX_ROW) && var2.startsWith(INTERAL_FN_INDEX_ROW))\n+ ||(var1.startsWith(INTERAL_FN_INDEX_COL) && var2.startsWith(INTERAL_FN_INDEX_COL)));\n}\nreturn ret;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "diff": "@@ -103,8 +103,52 @@ public class Connection implements Closeable\n* Connection constructor, the starting point for any other JMLC API calls.\n*\n*/\n- public Connection()\n- {\n+ public Connection() {\n+ //with default dml configuration\n+ this(new DMLConfig());\n+ }\n+\n+ /**\n+ * Connection constructor, the starting point for any other JMLC API calls.\n+ * This variant allows to enable a set of boolean compiler configurations.\n+ *\n+ * @param cconfigs one or many boolean compiler configurations to enable.\n+ */\n+ public Connection(CompilerConfig.ConfigType... cconfigs) {\n+ //basic constructor, which also constructs the compiler config\n+ this(new DMLConfig()); //with default dml configuration\n+\n+ //set optional compiler configurations in current config\n+ for( ConfigType configType : cconfigs )\n+ _cconf.set(configType, true);\n+ setLocalConfigs();\n+ }\n+\n+ /**\n+ * Connection constructor, the starting point for any other JMLC API calls.\n+ * This variant allows to pass a global dml configuration and enable a set\n+ * of boolean compiler configurations.\n+ *\n+ * @param dmlconfig a dml configuration.\n+ * @param cconfigs one or many boolean compiler configurations to enable.\n+ */\n+ public Connection(DMLConfig dmlconfig, CompilerConfig.ConfigType... cconfigs) {\n+ //basic constructor, which also constructs the compiler config\n+ this(dmlconfig);\n+\n+ //set optional compiler configurations in current config\n+ for( ConfigType configType : cconfigs )\n+ _cconf.set(configType, true);\n+ setLocalConfigs();\n+ }\n+\n+ /**\n+ * Connection constructor, the starting point for any other JMLC API calls.\n+ * This variant allows to pass a global dml configuration.\n+ *\n+ * @param dmlconfig a dml configuration.\n+ */\n+ public Connection(DMLConfig dmlconfig) {\nDMLScript.rtplatform = RUNTIME_PLATFORM.SINGLE_NODE;\n//setup basic parameters for embedded execution\n@@ -129,28 +173,12 @@ public class Connection implements Closeable\n//disable caching globally\nCacheableData.disableCaching();\n- //create default configuration\n- _dmlconf = new DMLConfig();\n+ //assign the given configuration\n+ _dmlconf = dmlconfig;\nsetLocalConfigs();\n}\n- /**\n- * Connection constructor, the starting point for any other JMLC API calls.\n- * This variant allows to enable a set of boolean compiler configurations.\n- *\n- * @param configs one or many boolean compiler configurations to enable.\n- */\n- public Connection(CompilerConfig.ConfigType... configs) {\n- //basic constructor, which also constructs the compiler config\n- this();\n-\n- //set optional compiler configurations in current config\n- for( ConfigType configType : configs )\n- _cconf.set(configType, true);\n- setLocalConfigs();\n- }\n-\n/**\n* Prepares (precompiles) a script and registers input and output variables.\n*\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ProgramBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ProgramBlock.java", "diff": "@@ -363,7 +363,13 @@ public class ProgramBlock implements ParseInfo\nsynchronized( mb ) { //potential state change\nmb.recomputeNonZeros();\nmb.examSparsity();\n+\n+ }\n+ if( mb.isInSparseFormat() && mb.isAllocated() ) {\n+ mb.getSparseBlock().checkValidity(mb.getNumRows(),\n+ mb.getNumColumns(), mb.getNonZeros(), true);\n}\n+\nboolean sparse2 = mb.isInSparseFormat();\nlong nnz2 = mb.getNonZeros();\nmo.release();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlock.java", "diff": "@@ -247,6 +247,20 @@ public abstract class SparseBlock implements Serializable\n*/\npublic abstract boolean isEmpty(int r);\n+ /**\n+ * Validate the correctness of the internal data structures of the different\n+ * sparse block implementations.\n+ *\n+ * @param rlen number of rows\n+ * @param clen number of columns\n+ * @param nnz number of non zeros\n+ * @param strict enforce optional properties\n+ * @return true if the sparse block is valid wrt the corresponding format\n+ * such as COO, CSR, MCSR.\n+ */\n+\n+ public abstract boolean checkValidity(int rlen, int clen, long nnz, boolean strict);\n+\n////////////////////////\n//obtain indexes/values/positions\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCOO.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCOO.java", "diff": "@@ -192,6 +192,12 @@ public class SparseBlockCOO extends SparseBlock\nreturn true;\n}\n+ @Override\n+ public boolean checkValidity(int rlen, int clen, long nnz, boolean strict) {\n+ //empty implementation\n+ return true;\n+ }\n+\n@Override\npublic void reset() {\n_size = 0;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "diff": "@@ -842,6 +842,57 @@ public class SparseBlockCSR extends SparseBlock\nreturn sb.toString();\n}\n+ @Override\n+ public boolean checkValidity(int rlen, int clen, long nnz, boolean strict) {\n+ //1. correct meta data\n+ if( rlen < 0 || clen < 0 ) {\n+ throw new RuntimeException(\"Invalid block dimensions: \"+rlen+\" \"+clen);\n+ }\n+\n+ //2. correct array lengths\n+ if(_size != nnz && _ptr.length < rlen+1 && _values.length < nnz && _indexes.length < nnz ) {\n+ throw new RuntimeException(\"Incorrect array lengths.\");\n+ }\n+\n+ //3. non-decreasing row pointers\n+ for( int i=1; i<rlen; i++ ) {\n+ if(_ptr[i-1] > _ptr[i] && strict)\n+ throw new RuntimeException(\"Row pointers are decreasing at row: \"+i\n+ + \", with pointers \"+_ptr[i-1]+\" > \"+_ptr[i]);\n+ }\n+\n+ //4. sorted column indexes per row\n+ for( int i=0; i<rlen; i++ ) {\n+ int apos = pos(i);\n+ int alen = size(i);\n+ for( int k=apos+1; k<apos+alen; k++)\n+ if( _indexes[k-1] >= _indexes[k] )\n+ throw new RuntimeException(\"Wrong sparse row ordering: \"\n+ + k + \" \"+_indexes[k-1]+\" \"+_indexes[k]);\n+ for( int k=apos; k<apos+alen; k++ )\n+ if( _values[k] == 0 )\n+ throw new RuntimeException(\"Wrong sparse row: zero at \"\n+ + k + \" at col index \" + _indexes[k]);\n+ }\n+\n+ //5. non-existing zero values\n+ for( int i=0; i<_size; i++ ) {\n+ if( _values[i] == 0 ) {\n+ throw new RuntimeException(\"The values array should not contain zeros.\"\n+ + \" The \" + i + \"th value is \"+_values[i]);\n+ }\n+ }\n+\n+ //6. a capacity that is no larger than nnz times resize factor.\n+ int capacity = _values.length;\n+ if(capacity > nnz*RESIZE_FACTOR1 ) {\n+ throw new RuntimeException(\"Capacity is larger than the nnz times a resize factor.\"\n+ + \" Current size: \"+capacity+ \", while Expected size:\"+nnz*RESIZE_FACTOR1);\n+ }\n+\n+ return true;\n+ }\n+\n///////////////////////////\n// private helper methods\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockMCSR.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockMCSR.java", "diff": "@@ -158,6 +158,12 @@ public class SparseBlockMCSR extends SparseBlock\nreturn (_rows[r] != null);\n}\n+ @Override\n+ public boolean checkValidity(int rlen, int clen, long nnz, boolean strict) {\n+ // empty implementation\n+ return true;\n+ }\n+\n@Override\npublic void reset() {\nfor( SparseRow row : _rows )\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/udf/ExternalFunctionInvocationInstruction.java", "new_path": "src/main/java/org/apache/sysml/udf/ExternalFunctionInvocationInstruction.java", "diff": "@@ -74,7 +74,7 @@ public class ExternalFunctionInvocationInstruction extends Instruction\nfun.setFunctionInputs(getInputObjects(inputs, ec.getVariables()));\n//executes function\n- fun.execute();\n+ fun.execute(ec);\n// get and verify the outputs\nverifyAndAttachOutputs(ec, fun, outputs);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/udf/PackageFunction.java", "new_path": "src/main/java/org/apache/sysml/udf/PackageFunction.java", "diff": "@@ -22,6 +22,7 @@ package org.apache.sysml.udf;\nimport java.io.Serializable;\nimport java.util.ArrayList;\n+import org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.controlprogram.parfor.util.IDSequence;\n/**\n@@ -159,4 +160,14 @@ public abstract class PackageFunction implements Serializable\n*/\npublic abstract void execute();\n+ /**\n+ * Method that will be executed to perform this function. The default\n+ * implementation simply forwards this call to execute.\n+ *\n+ * @param execution context with access to the program\n+ * e.g., for access to other dml-bodied or external functions.\n+ */\n+ public void execute(ExecutionContext ec) {\n+ execute(); //default impl\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/udf/PackageFunction.java", "new_path": "src/main/java/org/apache/sysml/udf/PackageFunction.java", "diff": "@@ -164,7 +164,7 @@ public abstract class PackageFunction implements Serializable\n* Method that will be executed to perform this function. The default\n* implementation simply forwards this call to execute.\n*\n- * @param execution context with access to the program\n+ * @param ec execution context with access to the program\n* e.g., for access to other dml-bodied or external functions.\n*/\npublic void execute(ExecutionContext ec) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "diff": "@@ -1038,6 +1038,7 @@ public class OptimizerUtils\n||(op==OpOp2.EQUAL && val!=0)\n||(op==OpOp2.MINUS && val==0)\n||(op==OpOp2.PLUS && val==0)\n+ ||(op==OpOp2.POW && val!=0)\n||(op==OpOp2.MAX && val<=0)\n||(op==OpOp2.MIN && val>=0));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "diff": "@@ -257,8 +257,10 @@ public class TemplateUtils\npublic static boolean rIsSparseSafeOnly(CNode node, BinType...types) {\nif( !(isBinary(node, types) || node instanceof CNodeData\n- || (node instanceof CNodeUnary && ((CNodeUnary)node).getType().isScalarLookup())\n- || (node instanceof CNodeUnary && ((CNodeUnary)node).getType().isSparseSafeScalar())) )\n+ || (node instanceof CNodeUnary && ((((CNodeUnary)node).getType().isScalarLookup())\n+ || ((CNodeUnary)node).getType().isSparseSafeScalar()\n+ || ((CNodeUnary)node).getType()==UnaryType.POW2\n+ || ((CNodeUnary)node).getType()==UnaryType.MULT2)) ))\nreturn false;\nboolean ret = true;\nfor( CNode c : node.getInput() )\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/mllearn/estimators.py", "new_path": "src/main/python/systemml/mllearn/estimators.py", "diff": "@@ -187,6 +187,18 @@ class BaseSystemMLEstimator(Estimator):\nself.y = None\nreturn self\n+ def fit_file(self, X_file, y_file):\n+ global default_jvm_stdout, default_jvm_stdout_parallel_flush\n+ try:\n+ if default_jvm_stdout:\n+ with jvm_stdout(parallel_flush=default_jvm_stdout_parallel_flush):\n+ self.model = self.estimator.fit(X_file, y_file)\n+ else:\n+ self.model = self.estimator.fit(X_file, y_file)\n+ except Py4JError:\n+ traceback.print_exc()\n+ return self\n+\n# Returns a model after calling fit(df) on Estimator object on JVM\ndef _fit(self, X):\n\"\"\"\n@@ -207,12 +219,14 @@ class BaseSystemMLEstimator(Estimator):\nParameters\n----------\n- X: NumPy ndarray, Pandas DataFrame, scipy sparse matrix\n- y: NumPy ndarray, Pandas DataFrame, scipy sparse matrix\n+ X: NumPy ndarray, Pandas DataFrame, scipy sparse matrix, Spark DataFrame, file path\n+ y: NumPy ndarray, Pandas DataFrame, scipy sparse matrix, file path\n\"\"\"\nif y is None:\nreturn self._fit(X)\n- elif y is not None and isinstance(X, SUPPORTED_TYPES) and isinstance(y, SUPPORTED_TYPES):\n+ elif isinstance(X, str) and isinstance(y, str):\n+ return self.fit_file(X, y)\n+ elif isinstance(X, SUPPORTED_TYPES) and isinstance(y, SUPPORTED_TYPES):\n# Donot encode if y is a numpy matrix => useful for segmentation\nskipEncodingY = len(y.shape) == 2 and y.shape[0] != 1 and y.shape[1] != 1\ny = y if skipEncodingY else self.encode(y)\n@@ -307,6 +321,8 @@ class BaseSystemMLEstimator(Estimator):\nexcept AttributeError:\npass\ntry:\n+ if isinstance(X, str):\n+ return self.model.transform_probability(X)\njX = self._convertPythonXToJavaObject(X)\nif default_jvm_stdout:\nwith jvm_stdout(parallel_flush=default_jvm_stdout_parallel_flush):\n@@ -323,7 +339,7 @@ class BaseSystemMLEstimator(Estimator):\nParameters\n----------\n- X: NumPy ndarray, Pandas DataFrame, scipy sparse matrix or PySpark DataFrame\n+ X: NumPy ndarray, Pandas DataFrame, scipy sparse matrix or PySpark DataFrame or file path\n\"\"\"\nglobal default_jvm_stdout, default_jvm_stdout_parallel_flush\ntry:\n@@ -332,6 +348,8 @@ class BaseSystemMLEstimator(Estimator):\nexcept AttributeError:\npass\ntry:\n+ if isinstance(X, str):\n+ return self.model.transform(X)\njX = self._convertPythonXToJavaObject(X)\nif default_jvm_stdout:\nwith jvm_stdout(parallel_flush=default_jvm_stdout_parallel_flush):\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/dl/Caffe2DML.scala", "new_path": "src/main/scala/org/apache/sysml/api/dl/Caffe2DML.scala", "diff": "@@ -206,6 +206,10 @@ class Caffe2DML(val sc: SparkContext,\nval that = new Caffe2DML(sc, solverParam, solver, net, lrPolicy, numChannels, height, width)\ncopyValues(that, extra)\n}\n+ def fit(X_file: String, y_file: String): Caffe2DMLModel = {\n+ mloutput = baseFit(X_file, y_file, sc)\n+ new Caffe2DMLModel(this)\n+ }\n// Note: will update the y_mb as this will be called by Python mllearn\ndef fit(X_mb: MatrixBlock, y_mb: MatrixBlock): Caffe2DMLModel = {\nmloutput = baseFit(X_mb, y_mb, sc)\n@@ -822,6 +826,15 @@ class Caffe2DMLModel(val numClasses: String, val sc: SparkContext, val solver: C\ndef baseEstimator(): BaseSystemMLEstimator = estimator\n// Prediction\n+ def transform(X_file: String): String =\n+ if (estimator.isClassification) {\n+ Caffe2DML.LOG.debug(\"Prediction assuming classification\")\n+ baseTransform(X_file, sc, \"Prob\")\n+ } else {\n+ Caffe2DML.LOG.debug(\"Prediction assuming segmentation\")\n+ val outShape = estimator.getOutputShapeOfLastLayer\n+ baseTransform(X_file, sc, \"Prob\", outShape._1.toInt, outShape._2.toInt, outShape._3.toInt)\n+ }\ndef transform(X: MatrixBlock): MatrixBlock =\nif (estimator.isClassification) {\nCaffe2DML.LOG.debug(\"Prediction assuming classification\")\n@@ -831,6 +844,15 @@ class Caffe2DMLModel(val numClasses: String, val sc: SparkContext, val solver: C\nval outShape = estimator.getOutputShapeOfLastLayer\nbaseTransform(X, sc, \"Prob\", outShape._1.toInt, outShape._2.toInt, outShape._3.toInt)\n}\n+ def transform_probability(X_file: String): String =\n+ if (estimator.isClassification) {\n+ Caffe2DML.LOG.debug(\"Prediction of probability assuming classification\")\n+ baseTransformProbability(X_file, sc, \"Prob\")\n+ } else {\n+ Caffe2DML.LOG.debug(\"Prediction of probability assuming segmentation\")\n+ val outShape = estimator.getOutputShapeOfLastLayer\n+ baseTransformProbability(X_file, sc, \"Prob\", outShape._1.toInt, outShape._2.toInt, outShape._3.toInt)\n+ }\ndef transform_probability(X: MatrixBlock): MatrixBlock =\nif (estimator.isClassification) {\nCaffe2DML.LOG.debug(\"Prediction of probability assuming classification\")\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLClassifier.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLClassifier.scala", "diff": "@@ -111,6 +111,11 @@ trait HasRegParam extends Params {\n}\ntrait BaseSystemMLEstimatorOrModel {\n+ def dmlRead(X:String, fileX:String):String = {\n+ val format = if(fileX.endsWith(\".csv\")) \", format=\\\"csv\\\"\" else \"\"\n+ return X + \" = read(\\\"\" + fileX + \"\\\"\" + format + \"); \"\n+ }\n+ def dmlWrite(X:String):String = \"write(\"+ X + \", \\\"output.mtx\\\", format=\\\"binary\\\"); \"\nvar enableGPU: Boolean = false\nvar forceGPU: Boolean = false\nvar explain: Boolean = false\n@@ -215,6 +220,16 @@ trait BaseSystemMLEstimatorModel extends BaseSystemMLEstimatorOrModel {\n}\ntrait BaseSystemMLClassifier extends BaseSystemMLEstimator {\n+ def baseFit(X_file: String, y_file: String, sc: SparkContext): MLResults = {\n+ val isSingleNode = false\n+ val ml = new MLContext(sc)\n+ updateML(ml)\n+ val readScript = dml(dmlRead(\"X\", X_file) + dmlRead(\"y\", y_file)).out(\"X\", \"y\")\n+ val res = ml.execute(readScript)\n+ val ret = getTrainingScript(isSingleNode)\n+ val script = ret._1.in(ret._2, res.getMatrix(\"X\")).in(ret._3, res.getMatrix(\"y\"))\n+ ml.execute(script)\n+ }\ndef baseFit(X_mb: MatrixBlock, y_mb: MatrixBlock, sc: SparkContext): MLResults = {\nval isSingleNode = true\nval ml = new MLContext(sc)\n@@ -242,8 +257,33 @@ trait BaseSystemMLClassifier extends BaseSystemMLEstimator {\ntrait BaseSystemMLClassifierModel extends BaseSystemMLEstimatorModel {\n+ def baseTransform(X_file: String, sc: SparkContext, probVar: String): String = baseTransform(X_file, sc, probVar, -1, 1, 1)\ndef baseTransform(X: MatrixBlock, sc: SparkContext, probVar: String): MatrixBlock = baseTransform(X, sc, probVar, -1, 1, 1)\n+ def baseTransform(X: String, sc: SparkContext, probVar: String, C: Int, H: Int, W: Int): String = {\n+ val Prob = baseTransformHelper(X, sc, probVar, C, H, W)\n+ val script1 = dml(\"source(\\\"nn/util.dml\\\") as util; Prediction = util::predict_class(Prob, C, H, W); \" + dmlWrite(\"Prediction\"))\n+ .in(\"Prob\", Prob)\n+ .in(\"C\", C)\n+ .in(\"H\", H)\n+ .in(\"W\", W)\n+ val ml = new MLContext(sc)\n+ updateML(ml)\n+ ml.execute(script1)\n+ return \"output.mtx\"\n+ }\n+\n+ def baseTransformHelper(X_file: String, sc: SparkContext, probVar: String, C: Int, H: Int, W: Int): Matrix = {\n+ val isSingleNode = true\n+ val ml = new MLContext(sc)\n+ updateML(ml)\n+ val readScript = dml(dmlRead(\"X\", X_file)).out(\"X\")\n+ val res = ml.execute(readScript)\n+ val script = getPredictionScript(isSingleNode)\n+ val modelPredict = ml.execute(script._1.in(script._2, res.getMatrix(\"X\")))\n+ return modelPredict.getMatrix(probVar)\n+ }\n+\ndef baseTransform(X: MatrixBlock, sc: SparkContext, probVar: String, C: Int, H: Int, W: Int): MatrixBlock = {\nval Prob = baseTransformHelper(X, sc, probVar, C, H, W)\nval script1 = dml(\"source(\\\"nn/util.dml\\\") as util; Prediction = util::predict_class(Prob, C, H, W);\")\n@@ -282,9 +322,18 @@ trait BaseSystemMLClassifierModel extends BaseSystemMLEstimatorModel {\ndef baseTransformProbability(X: MatrixBlock, sc: SparkContext, probVar: String): MatrixBlock =\nbaseTransformProbability(X, sc, probVar, -1, 1, 1)\n+ def baseTransformProbability(X: String, sc: SparkContext, probVar: String): String =\n+ baseTransformProbability(X, sc, probVar, -1, 1, 1)\n+\ndef baseTransformProbability(X: MatrixBlock, sc: SparkContext, probVar: String, C: Int, H: Int, W: Int): MatrixBlock =\nreturn baseTransformHelper(X, sc, probVar, C, H, W).toMatrixBlock\n+ def baseTransformProbability(X: String, sc: SparkContext, probVar: String, C: Int, H: Int, W: Int): String = {\n+ val Prob = baseTransformHelper(X, sc, probVar, C, H, W)\n+ (new MLContext(sc)).execute(dml(dmlWrite(\"Prob\")).in(\"Prob\", Prob))\n+ \"output.mtx\"\n+ }\n+\ndef baseTransform(df: ScriptsUtils.SparkDataType, sc: SparkContext, probVar: String, outputProb: Boolean = true): DataFrame =\nbaseTransform(df, sc, probVar, outputProb, -1, 1, 1)\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLRegressor.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLRegressor.scala", "diff": "@@ -35,6 +35,17 @@ import org.apache.sysml.api.mlcontext.ScriptFactory._\ntrait BaseSystemMLRegressor extends BaseSystemMLEstimator {\n+ def baseFit(X_file: String, y_file: String, sc: SparkContext): MLResults = {\n+ val isSingleNode = false\n+ val ml = new MLContext(sc)\n+ updateML(ml)\n+ val readScript = dml(dmlRead(\"X\", X_file) + dmlRead(\"y\", y_file)).out(\"X\", \"y\")\n+ val res = ml.execute(readScript)\n+ val ret = getTrainingScript(isSingleNode)\n+ val script = ret._1.in(ret._2, res.getMatrix(\"X\")).in(ret._3, res.getMatrix(\"y\"))\n+ ml.execute(script)\n+ }\n+\ndef baseFit(X_mb: MatrixBlock, y_mb: MatrixBlock, sc: SparkContext): MLResults = {\nval isSingleNode = true\nval ml = new MLContext(sc)\n@@ -61,6 +72,18 @@ trait BaseSystemMLRegressor extends BaseSystemMLEstimator {\ntrait BaseSystemMLRegressorModel extends BaseSystemMLEstimatorModel {\n+ def baseTransform(X_file: String, sc: SparkContext, predictionVar: String): String = {\n+ val isSingleNode = false\n+ val ml = new MLContext(sc)\n+ updateML(ml)\n+ val readScript = dml(dmlRead(\"X\", X_file)).out(\"X\")\n+ val res = ml.execute(readScript)\n+ val script = getPredictionScript(isSingleNode)\n+ val modelPredict = ml.execute(script._1.in(script._2, res.getMatrix(\"X\")))\n+ val writeScript = dml(dmlWrite(\"X\")).in(\"X\", modelPredict.getMatrix(predictionVar))\n+ ml.execute(writeScript)\n+ return \"output.mtx\"\n+ }\ndef baseTransform(X: MatrixBlock, sc: SparkContext, predictionVar: String): MatrixBlock = {\nval isSingleNode = true\nval ml = new MLContext(sc)\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/LinearRegression.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/LinearRegression.scala", "diff": "@@ -77,6 +77,11 @@ class LinearRegression(override val uid: String, val sc: SparkContext, val solve\n(script, \"X\", \"y\")\n}\n+ def fit(X_file: String, y_file: String): LinearRegressionModel = {\n+ mloutput = baseFit(X_file, y_file, sc)\n+ new LinearRegressionModel(this)\n+ }\n+\ndef fit(X_mb: MatrixBlock, y_mb: MatrixBlock): LinearRegressionModel = {\nmloutput = baseFit(X_mb, y_mb, sc)\nnew LinearRegressionModel(this)\n@@ -102,6 +107,7 @@ class LinearRegressionModel(override val uid: String)(estimator: LinearRegressio\n}\ndef transform_probability(X: MatrixBlock): MatrixBlock = throw new DMLRuntimeException(\"Unsupported method\")\n+ def transform_probability(X_file: String): String = throw new DMLRuntimeException(\"Unsupported method\")\ndef baseEstimator(): BaseSystemMLEstimator = estimator\n@@ -115,7 +121,6 @@ class LinearRegressionModel(override val uid: String)(estimator: LinearRegressio\ndef modelVariables(): List[String] = List[String](\"beta_out\")\ndef transform(df: ScriptsUtils.SparkDataType): DataFrame = baseTransform(df, sc, \"means\")\n-\ndef transform(X: MatrixBlock): MatrixBlock = baseTransform(X, sc, \"means\")\n-\n+ def transform(X_file: String): String = baseTransform(X_file, sc, \"means\")\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/LogisticRegression.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/LogisticRegression.scala", "diff": "@@ -60,6 +60,11 @@ class LogisticRegression(override val uid: String, val sc: SparkContext)\ncopyValues(that, extra)\n}\n+ def fit(X_file: String, y_file: String): LogisticRegressionModel = {\n+ mloutput = baseFit(X_file, y_file, sc)\n+ new LogisticRegressionModel(this)\n+ }\n+\n// Note: will update the y_mb as this will be called by Python mllearn\ndef fit(X_mb: MatrixBlock, y_mb: MatrixBlock): LogisticRegressionModel = {\nmloutput = baseFit(X_mb, y_mb, sc)\n@@ -116,7 +121,9 @@ class LogisticRegressionModel(override val uid: String)(estimator: LogisticRegre\ndef modelVariables(): List[String] = List[String](\"B_out\")\ndef transform(X: MatrixBlock): MatrixBlock = baseTransform(X, sc, \"means\")\n+ def transform(X: String): String = baseTransform(X, sc, \"means\")\ndef transform_probability(X: MatrixBlock): MatrixBlock = baseTransformProbability(X, sc, \"means\")\n+ def transform_probability(X: String): String = baseTransformProbability(X, sc, \"means\")\ndef transform(df: ScriptsUtils.SparkDataType): DataFrame = baseTransform(df, sc, \"means\")\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/NaiveBayes.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/NaiveBayes.scala", "diff": "@@ -44,6 +44,11 @@ class NaiveBayes(override val uid: String, val sc: SparkContext) extends Estimat\n}\ndef setLaplace(value: Double) = set(laplace, value)\n+ def fit(X_file: String, y_file: String): NaiveBayesModel = {\n+ mloutput = baseFit(X_file, y_file, sc)\n+ new NaiveBayesModel(this)\n+ }\n+\n// Note: will update the y_mb as this will be called by Python mllearn\ndef fit(X_mb: MatrixBlock, y_mb: MatrixBlock): NaiveBayesModel = {\nmloutput = baseFit(X_mb, y_mb, sc)\n@@ -108,7 +113,9 @@ class NaiveBayesModel(override val uid: String)(estimator: NaiveBayes, val sc: S\ndef baseEstimator(): BaseSystemMLEstimator = estimator\ndef transform(X: MatrixBlock): MatrixBlock = baseTransform(X, sc, \"probs\")\n+ def transform(X: String): String = baseTransform(X, sc, \"probs\")\ndef transform_probability(X: MatrixBlock): MatrixBlock = baseTransformProbability(X, sc, \"probs\")\n+ def transform_probability(X: String): String = baseTransformProbability(X, sc, \"probs\")\ndef transform(df: ScriptsUtils.SparkDataType): DataFrame = baseTransform(df, sc, \"probs\")\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/SVM.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/SVM.scala", "diff": "@@ -70,6 +70,11 @@ class SVM(override val uid: String, val sc: SparkContext, val isMultiClass: Bool\n(script, \"X\", \"Y\")\n}\n+ def fit(X_file: String, y_file: String): SVMModel = {\n+ mloutput = baseFit(X_file, y_file, sc)\n+ new SVMModel(this, isMultiClass)\n+ }\n+\n// Note: will update the y_mb as this will be called by Python mllearn\ndef fit(X_mb: MatrixBlock, y_mb: MatrixBlock): SVMModel = {\nmloutput = baseFit(X_mb, y_mb, sc)\n@@ -121,5 +126,7 @@ class SVMModel(override val uid: String)(estimator: SVM, val sc: SparkContext, v\ndef transform(X: MatrixBlock): MatrixBlock = baseTransform(X, sc, \"scores\")\ndef transform_probability(X: MatrixBlock): MatrixBlock = baseTransformProbability(X, sc, \"scores\")\n+ def transform(X: String): String = baseTransform(X, sc, \"scores\")\n+ def transform_probability(X: String): String = baseTransformProbability(X, sc, \"scores\")\ndef transform(df: ScriptsUtils.SparkDataType): DataFrame = baseTransform(df, sc, \"scores\")\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCOO.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCOO.java", "diff": "@@ -194,7 +194,50 @@ public class SparseBlockCOO extends SparseBlock\n@Override\npublic boolean checkValidity(int rlen, int clen, long nnz, boolean strict) {\n- //empty implementation\n+ //1. correct meta data\n+ if(rlen < 0 || clen < 0) {\n+ throw new RuntimeException(\"Invalid block dimensions: \"+rlen+\" \"+clen);\n+ }\n+\n+ //2. correct array lengths\n+ if(_size != nnz && _cindexes.length < nnz && _rindexes.length < nnz && _values.length < nnz) {\n+ throw new RuntimeException(\"Incorrect array lengths.\");\n+ }\n+\n+ //3.1. sort order of row indices\n+ for( int i=1; i<=nnz; i++ ) {\n+ if(_rindexes[i] < _rindexes[i-1])\n+ throw new RuntimeException(\"Wrong sorted order of row indices\");\n+ }\n+\n+ //3.2. sorted values wrt to col indexes wrt to a given row index\n+ for( int i=0; i<rlen; i++ ) {\n+ int apos = pos(i);\n+ int alen = size(i);\n+ for(int k=apos+i; k<apos+alen; k++)\n+ if( _cindexes[k+1] >= _cindexes[k] )\n+ throw new RuntimeException(\"Wrong sparse row ordering: \"\n+ + k + \" \"+_cindexes[k-1]+\" \"+_cindexes[k]);\n+ for( int k=apos; k<apos+alen; k++ )\n+ if(_values[k] == 0)\n+ throw new RuntimeException(\"Wrong sparse row: zero at \"\n+ + k + \" at col index \" + _cindexes[k]);\n+ }\n+\n+ //4. non-existing zero values\n+ for( int i=0; i<_size; i++ ) {\n+ if( _values[i] == 0)\n+ throw new RuntimeException(\"The values array should not contain zeros.\"\n+ + \" The \" + i + \"th value is \"+_values[i]);\n+ }\n+\n+ //5. a capacity that is no larger than nnz times the resize factor\n+ int capacity = _values.length;\n+ if( capacity > nnz*RESIZE_FACTOR1 ) {\n+ throw new RuntimeException(\"Capacity is larger than the nnz times a resize factor.\"\n+ + \" Current size: \"+capacity+ \", while Expected size:\"+nnz*RESIZE_FACTOR1);\n+ }\n+\nreturn true;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/InvalidFunctionAssignmentTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/InvalidFunctionAssignmentTest.java", "diff": "@@ -25,9 +25,6 @@ import org.apache.sysml.api.DMLException;\nimport org.apache.sysml.test.integration.AutomatedTestBase;\nimport org.apache.sysml.test.integration.TestConfiguration;\n-/**\n- *\n- */\npublic class InvalidFunctionAssignmentTest extends AutomatedTestBase\n{\nprivate final static String TEST_DIR = \"functions/misc/\";\n@@ -52,7 +49,8 @@ public class InvalidFunctionAssignmentTest extends AutomatedTestBase\n@Test\npublic void testInvalidFunctionAssignmentInlined() {\n- runTest( TEST_NAME2, true );\n+ //MB: we now support UDFs in expressions and hence it's valid\n+ runTest( TEST_NAME2, false );\n}\n@Test\n@@ -62,23 +60,16 @@ public class InvalidFunctionAssignmentTest extends AutomatedTestBase\n@Test\npublic void testInvalidFunctionAssignment() {\n- runTest( TEST_NAME4, true );\n+ //MB: we now support UDFs in expressions and hence it's valid\n+ runTest( TEST_NAME4, false );\n}\n- /**\n- *\n- * @param testName\n- */\n- private void runTest( String testName, boolean exceptionExpected )\n- {\n+ private void runTest( String testName, boolean exceptionExpected ) {\nTestConfiguration config = getTestConfiguration(testName);\nloadTestConfiguration(config);\n-\nString HOME = SCRIPT_DIR + TEST_DIR;\nfullDMLScriptName = HOME + testName + \".dml\";\nprogramArgs = new String[]{};\n-\n- //run tests\nrunTest(true, exceptionExpected, DMLException.class, -1);\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/FunctionInExpressionTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/FunctionInExpressionTest.java", "diff": "@@ -33,6 +33,8 @@ public class FunctionInExpressionTest extends AutomatedTestBase\nprivate final static String TEST_NAME2 = \"FunInExpression2\";\nprivate final static String TEST_NAME3 = \"FunInExpression3\";\nprivate final static String TEST_NAME4 = \"FunInExpression4\";\n+ private final static String TEST_NAME5 = \"FunInExpression5\";\n+ private final static String TEST_NAME6 = \"FunInExpression6\";\nprivate final static String TEST_DIR = \"functions/misc/\";\nprivate final static String TEST_CLASS_DIR = TEST_DIR + FunctionInExpressionTest.class.getSimpleName() + \"/\";\n@@ -44,6 +46,8 @@ public class FunctionInExpressionTest extends AutomatedTestBase\naddTestConfiguration( TEST_NAME2, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME2, new String[] { \"R\" }) );\naddTestConfiguration( TEST_NAME3, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME3, new String[] { \"R\" }) );\naddTestConfiguration( TEST_NAME4, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME4, new String[] { \"R\" }) );\n+ addTestConfiguration( TEST_NAME5, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME5, new String[] { \"R\" }) );\n+ addTestConfiguration( TEST_NAME6, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME6, new String[] { \"R\" }) );\n}\n@Test\n@@ -66,6 +70,16 @@ public class FunctionInExpressionTest extends AutomatedTestBase\nrunFunInExpressionTest( TEST_NAME4 );\n}\n+ @Test\n+ public void testFunInExpression5() {\n+ runFunInExpressionTest( TEST_NAME5 );\n+ }\n+\n+ @Test\n+ public void testFunInExpression6() {\n+ runFunInExpressionTest( TEST_NAME6 );\n+ }\n+\nprivate void runFunInExpressionTest( String testName )\n{\nTestConfiguration config = getTestConfiguration(testName);\n@@ -83,6 +97,6 @@ public class FunctionInExpressionTest extends AutomatedTestBase\n//compare results\ndouble val = readDMLMatrixFromHDFS(\"R\").get(new CellIndex(1,1));\n- Assert.assertTrue(\"Wrong result: 7 vs \"+val, Math.abs(val-7)<Math.pow(10, -14));\n+ Assert.assertTrue(\"Wrong result: 7 vs \"+val, Math.abs(val-7)<Math.pow(10, -13));\n}\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/FunInExpression5.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+\n+orderExternal = externalFunction(Matrix[Double] A, Integer col, Boolean desc) return (Matrix[Double] B)\n+ implemented in (classname=\"org.apache.sysml.udf.lib.OrderWrapper\",exectype=\"mem\")\n+\n+foo = function( Matrix[Double] A ) return (Matrix[Double] B) {\n+ for( i in 1:ncol(A) ) {\n+ B = orderExternal(A, i, TRUE);\n+ }\n+}\n+\n+A = matrix( 0.07, rows=10, cols=10 );\n+R = foo(A) * foo(A) + 7;\n+R = as.matrix( sum( sqrt(R-7) ) );\n+\n+write( R, $1 ); #ordered input\n\\ No newline at end of file\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/FunInExpression6.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+\n+orderExternal = externalFunction(Matrix[Double] A, Integer col, Boolean desc) return (Matrix[Double] B)\n+ implemented in (classname=\"org.apache.sysml.udf.lib.OrderWrapper\",exectype=\"mem\")\n+\n+A = matrix( 0.07, rows=10, cols=10 );\n+R = orderExternal(A*A, 7, TRUE) + 7;\n+R = as.matrix( sum( sqrt(R-7) ) );\n+\n+write( R, $1 ); #ordered input\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/common/CommonSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/common/CommonSyntacticValidator.java", "diff": "@@ -573,63 +573,51 @@ public abstract class CommonSyntacticValidator {\n*/\nprotected abstract Expression handleLanguageSpecificFunction(ParserRuleContext ctx, String functionName, ArrayList<ParameterExpression> paramExpressions);\n- /** Checks for builtin functions and does Action 'f'.\n- * <p>\n- * Constructs the\n- * appropriate {@link AssignmentStatement} from\n- * {@link CommonSyntacticValidator#functionCallAssignmentStatementHelper(ParserRuleContext, Set, Set, Expression, StatementInfo, Token, Token, String, String, ArrayList, boolean)}\n- * or Assign to {@link Expression} from\n- * DmlSyntacticValidator's exitBuiltinFunctionExpression(BuiltinFunctionExpressionContext).\n- * </p>\n+ /** Creates a builtin function expression.\n*\n* @param ctx antlr rule context\n* @param functionName Name of the builtin function\n* @param paramExpressions Array of parameter names and values\n- * @param f action to perform\n- * @return true if a builtin function was found\n+ * @return expression if found otherwise null\n*/\n- protected boolean buildForBuiltInFunction(ParserRuleContext ctx, String functionName, ArrayList<ParameterExpression> paramExpressions, Action f) {\n+ protected Expression buildForBuiltInFunction(ParserRuleContext ctx, String functionName, ArrayList<ParameterExpression> paramExpressions) {\n// In global namespace, so it can be a builtin function\n// Double verification: verify passed function name is a (non-parameterized) built-in function.\ntry {\nif (functions.contains(functionName)) {\n// It is a user function definition (which takes precedence if name same as built-in)\n- return false;\n+ return null;\n}\nExpression lsf = handleLanguageSpecificFunction(ctx, functionName, paramExpressions);\nif (lsf != null) {\nsetFileLineColumn(lsf, ctx);\n- f.execute(lsf);\n- return true;\n+ return lsf;\n}\nBuiltinFunctionExpression bife = BuiltinFunctionExpression.getBuiltinFunctionExpression(ctx, functionName, paramExpressions, currentFile);\nif (bife != null) {\n// It is a builtin function\n- f.execute(bife);\n- return true;\n+ return bife;\n}\nParameterizedBuiltinFunctionExpression pbife = ParameterizedBuiltinFunctionExpression\n.getParamBuiltinFunctionExpression(ctx, functionName, paramExpressions, currentFile);\nif (pbife != null){\n// It is a parameterized builtin function\n- f.execute(pbife);\n- return true;\n+ return pbife;\n}\n// built-in read, rand ...\nDataExpression dbife = DataExpression.getDataExpression(ctx, functionName, paramExpressions, currentFile, errorListener);\nif (dbife != null){\n- f.execute(dbife);\n- return true;\n+ return dbife;\n+ }\n}\n- } catch(Exception e) {\n+ catch(Exception e) {\nnotifyErrorListeners(\"unable to process builtin function expression \" + functionName + \":\" + e.getMessage(), ctx.start);\n- return true;\n}\n- return false;\n+ return null;\n}\n@@ -670,14 +658,12 @@ public abstract class CommonSyntacticValidator {\n// For builtin functions with LHS\nif(namespace.equals(DMLProgram.DEFAULT_NAMESPACE) && !functions.contains(functionName)){\n- final DataIdentifier ftarget = target;\n- Action f = new Action() {\n- @Override public void execute(Expression e) { setAssignmentStatement(ctx, info , ftarget, e); }\n- };\n- boolean validBIF = buildForBuiltInFunction(ctx, functionName, paramExpression, f);\n- if (validBIF)\n+ Expression e = buildForBuiltInFunction(ctx, functionName, paramExpression);\n+ if( e != null ) {\n+ setAssignmentStatement(ctx, info, target, e);\nreturn;\n}\n+ }\n// handle user-defined functions\nsetAssignmentStatement(ctx, info, target,\n@@ -695,14 +681,6 @@ public abstract class CommonSyntacticValidator {\nreturn functCall;\n}\n- /**\n- * To allow for different actions in\n- * {@link CommonSyntacticValidator#functionCallAssignmentStatementHelper(ParserRuleContext, Set, Set, Expression, StatementInfo, Token, Token, String, String, ArrayList, boolean)}\n- */\n- public static interface Action {\n- public void execute(Expression e);\n- }\n-\nprotected void setMultiAssignmentStatement(ArrayList<DataIdentifier> target, Expression expression, ParserRuleContext ctx, StatementInfo info) {\ninfo.stmt = new MultiAssignmentStatement(target, expression);\ninfo.stmt.setCtxValuesAndFilename(ctx, currentFile);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "diff": "@@ -530,18 +530,14 @@ public class DmlSyntacticValidator extends CommonSyntacticValidator implements D\nfunctionName = convertedSyntax.functionName;\nparamExpression = convertedSyntax.paramExpression;\n}\n- final ExpressionInfo info = ctx.info;\n- Action f = new Action() {\n- @Override public void execute(Expression e) { info.expr = e; }\n- };\n// handle built-in functions\n- boolean validBIF = buildForBuiltInFunction(ctx, functionName, paramExpression, f);\n- if (validBIF)\n+ ctx.info.expr = buildForBuiltInFunction(ctx, functionName, paramExpression);\n+ if( ctx.info.expr != null )\nreturn;\n// handle user-defined functions\n- info.expr = createFunctionCall(ctx, namespace, functionName, paramExpression);\n+ ctx.info.expr = createFunctionCall(ctx, namespace, functionName, paramExpression);\n}\n@@ -583,14 +579,12 @@ public class DmlSyntacticValidator extends CommonSyntacticValidator implements D\n}\nif(namespace.equals(DMLProgram.DEFAULT_NAMESPACE)) {\n- final FunctionCallMultiAssignmentStatementContext fctx = ctx;\n- Action f = new Action() {\n- @Override public void execute(Expression e) { setMultiAssignmentStatement(targetList, e, fctx, fctx.info); }\n- };\n- boolean validBIF = buildForBuiltInFunction(ctx, functionName, paramExpression, f);\n- if (validBIF)\n+ Expression e = buildForBuiltInFunction(ctx, functionName, paramExpression);\n+ if( e != null ) {\n+ setMultiAssignmentStatement(targetList, e, ctx, ctx.info);\nreturn;\n}\n+ }\n// Override default namespace for imported non-built-in function\nString inferNamespace = (sourceNamespace != null && sourceNamespace.length() > 0 && DMLProgram.DEFAULT_NAMESPACE.equals(namespace)) ? sourceNamespace : namespace;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/pydml/PydmlSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/pydml/PydmlSyntacticValidator.java", "diff": "@@ -1145,18 +1145,13 @@ public class PydmlSyntacticValidator extends CommonSyntacticValidator implements\nparamExpression = convertedSyntax.paramExpression;\n}\n- final ExpressionInfo info = ctx.info;\n- Action f = new Action() {\n- @Override public void execute(Expression e) { info.expr = e; }\n- };\n-\n//handle builtin functions\n- boolean validBIF = buildForBuiltInFunction(ctx, functionName, paramExpression, f);\n- if (validBIF)\n+ ctx.info.expr= buildForBuiltInFunction(ctx, functionName, paramExpression);\n+ if( ctx.info.expr != null )\nreturn;\n// handle user-defined functions\n- info.expr = createFunctionCall(ctx, namespace, functionName, paramExpression);\n+ ctx.info.expr = createFunctionCall(ctx, namespace, functionName, paramExpression);\n}\n@Override\n@@ -1199,14 +1194,12 @@ public class PydmlSyntacticValidator extends CommonSyntacticValidator implements\n}\nif(namespace.equals(DMLProgram.DEFAULT_NAMESPACE)) {\n- final FunctionCallMultiAssignmentStatementContext fctx = ctx;\n- Action f = new Action() {\n- @Override public void execute(Expression e) { setMultiAssignmentStatement(targetList, e, fctx, fctx.info); }\n- };\n- boolean validBIF = buildForBuiltInFunction(ctx, functionName, paramExpression, f);\n- if (validBIF)\n+ Expression e = buildForBuiltInFunction(ctx, functionName, paramExpression);\n+ if( e != null ) {\n+ setMultiAssignmentStatement(targetList, e, ctx, ctx.info);\nreturn;\n}\n+ }\n// Override default namespace for imported non-built-in function\nString inferNamespace = (sourceNamespace != null && sourceNamespace.length() > 0 && DMLProgram.DEFAULT_NAMESPACE.equals(namespace)) ? sourceNamespace : namespace;\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/nn/layers/lstm.dml", "new_path": "scripts/nn/layers/lstm.dml", "diff": "@@ -90,15 +90,12 @@ forward = function(matrix[double] X, matrix[double] W, matrix[double] b, int T,\nX_t = X[,(t-1)*D+1:t*D] # shape (N, D)\ninput = cbind(X_t, out_prev) # shape (N, D+M)\nifog = input %*% W + b # input, forget, output, and g gates; shape (N, 4M)\n- tmp = sigmoid::forward(ifog[,1:3*M]) # i,f,o gates squashed with sigmoid\n- ifog[,1:3*M] = tmp\n- tmp = tanh::forward(ifog[,3*M+1:4*M]) # g gate squashed with tanh\n- ifog[,3*M+1:4*M] = tmp\n+ ifog[,1:3*M] = sigmoid::forward(ifog[,1:3*M]) # i,f,o gates squashed with sigmoid\n+ ifog[,3*M+1:4*M] = tanh::forward(ifog[,3*M+1:4*M]) # g gate squashed with tanh\n# c_t = f*prev_c + i*g\nc = ifog[,M+1:2*M]*c_prev + ifog[,1:M]*ifog[,3*M+1:4*M] # shape (N, M)\n# out_t = o*tanh(c)\n- tmp = tanh::forward(c)\n- out_t = ifog[,2*M+1:3*M] * tmp # shape (N, M)\n+ out_t = ifog[,2*M+1:3*M] * tanh::forward(c) # shape (N, M)\n# store\nif (return_sequences) {\n@@ -202,10 +199,8 @@ backward = function(matrix[double] dout, matrix[double] dc,\no = ifog[,2*M+1:3*M] # output gate, shape (N, M)\ng = ifog[,3*M+1:4*M] # g gate, shape (N, M)\n- tmp = tanh::backward(dout_t, ct)\n- dct = dct + o*tmp # shape (N, M)\n- tmp = tanh::forward(ct)\n- do = tmp * dout_t # output gate, shape (N, M)\n+ dct = dct + o*tanh::backward(dout_t, ct) # shape (N, M)\n+ do = tanh::forward(ct) * dout_t # output gate, shape (N, M)\ndf = c_prev * dct # forget gate, shape (N, M)\ndc_prev = f * dct # shape (N, M)\ndi = g * dct # input gate, shape (N, M)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/JMLCInputOutputTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/JMLCInputOutputTest.java", "diff": "@@ -108,23 +108,21 @@ public class JMLCInputOutputTest extends AutomatedTestBase {\nconn.close();\n}\n- // See: https://issues.apache.org/jira/browse/SYSTEMML-658\n- // @Test\n- // public void testScalarInputString() throws IOException, DMLException {\n- // Connection conn = new Connection();\n- // String str = conn.readScript(baseDirectory + File.separator +\n- // \"scalar-input.dml\");\n- // PreparedScript script = conn.prepareScript(str, new String[] {\n- // \"inScalar1\", \"inScalar2\" }, new String[] {},\n- // false);\n- // String inScalar1 = \"hello\";\n- // String inScalar2 = \"goodbye\";\n- // script.setScalar(\"inScalar1\", inScalar1);\n- // script.setScalar(\"inScalar2\", inScalar2);\n- //\n- // setExpectedStdOut(\"total:hellogoodbye\");\n- // script.executeScript();\n- // }\n+ @Test\n+ public void testScalarInputString() throws IOException, DMLException {\n+ Connection conn = new Connection();\n+ String str = conn.readScript(baseDirectory + File.separator + \"scalar-input.dml\");\n+ PreparedScript script = conn.prepareScript(str, new String[] { \"inScalar1\", \"inScalar2\" }, new String[] {},\n+ false);\n+ String inScalar1 = \"Plant\";\n+ String inScalar2 = \" Trees\";\n+ script.setScalar(\"inScalar1\", inScalar1);\n+ script.setScalar(\"inScalar2\", inScalar2);\n+\n+ setExpectedStdOut(\"total:Plant Trees\");\n+ script.executeScript();\n+ conn.close();\n+ }\n@Test\npublic void testScalarInputStringExplicitValueType() throws IOException, DMLException {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "diff": "@@ -34,6 +34,8 @@ import org.apache.sysml.runtime.instructions.InstructionUtils;\nimport org.apache.sysml.runtime.instructions.cp.CPOperand;\nimport org.apache.sysml.runtime.instructions.spark.functions.FilterNonEmptyBlocksFunction;\nimport org.apache.sysml.runtime.instructions.spark.utils.RDDAggregateUtils;\n+import org.apache.sysml.runtime.instructions.spark.utils.SparkUtils;\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;\n@@ -85,17 +87,24 @@ public class CpmmSPInstruction extends BinarySPInstruction {\n//get rdd inputs\nJavaPairRDD<MatrixIndexes,MatrixBlock> in1 = sec.getBinaryBlockRDDHandleForVariable(input1.getName());\nJavaPairRDD<MatrixIndexes,MatrixBlock> in2 = sec.getBinaryBlockRDDHandleForVariable(input2.getName());\n+ MatrixCharacteristics mc1 = sec.getMatrixCharacteristics(input1.getName());\n+ MatrixCharacteristics mc2 = sec.getMatrixCharacteristics(input2.getName());\n+\nif( _aggtype == SparkAggType.SINGLE_BLOCK ) {\n//prune empty blocks of ultra-sparse matrices\nin1 = in1.filter(new FilterNonEmptyBlocksFunction());\nin2 = in2.filter(new FilterNonEmptyBlocksFunction());\n}\n+ //compute preferred join degree of parallelism\n+ int numPreferred = getPreferredParJoin(mc1, mc2, in1.getNumPartitions(), in2.getNumPartitions());\n+ int numPartJoin = Math.min(getMaxParJoin(mc1, mc2), numPreferred);\n+\n//process core cpmm matrix multiply\nJavaPairRDD<Long, IndexedMatrixValue> tmp1 = in1.mapToPair(new CpmmIndexFunction(true));\nJavaPairRDD<Long, IndexedMatrixValue> tmp2 = in2.mapToPair(new CpmmIndexFunction(false));\nJavaPairRDD<MatrixIndexes,MatrixBlock> out = tmp1\n- .join(tmp2) // join over common dimension\n+ .join(tmp2, numPartJoin) // join over common dimension\n.mapToPair(new CpmmMultiplyFunction()); // compute block multiplications\n//process cpmm aggregation and handle outputs\n@@ -121,6 +130,22 @@ public class CpmmSPInstruction extends BinarySPInstruction {\n}\n}\n+ private static int getPreferredParJoin(MatrixCharacteristics mc1, MatrixCharacteristics mc2, int numPar1, int numPar2) {\n+ int defPar = SparkExecutionContext.getDefaultParallelism(true);\n+ int maxParIn = Math.max(numPar1, numPar2);\n+ int maxSizeIn = SparkUtils.getNumPreferredPartitions(mc1) +\n+ SparkUtils.getNumPreferredPartitions(mc2);\n+ int tmp = (mc1.dimsKnown(true) && mc2.dimsKnown(true)) ?\n+ Math.max(maxSizeIn, maxParIn) : maxParIn;\n+ return (tmp > defPar/2) ? Math.max(tmp, defPar) : tmp;\n+ }\n+\n+ private static int getMaxParJoin(MatrixCharacteristics mc1, MatrixCharacteristics mc2) {\n+ return mc1.colsKnown() ? (int)mc1.getNumColBlocks() :\n+ mc2.rowsKnown() ? (int)mc2.getNumRowBlocks() :\n+ Integer.MAX_VALUE;\n+ }\n+\nprivate static class CpmmIndexFunction implements PairFunction<Tuple2<MatrixIndexes, MatrixBlock>, Long, IndexedMatrixValue>\n{\nprivate static final long serialVersionUID = -1187183128301671162L;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockMCSR.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockMCSR.java", "diff": "@@ -160,7 +160,38 @@ public class SparseBlockMCSR extends SparseBlock\n@Override\npublic boolean checkValidity(int rlen, int clen, long nnz, boolean strict) {\n- // empty implementation\n+\n+ //1. Correct meta data\n+ if( rlen < 0 || clen < 0 )\n+ throw new RuntimeException(\"Invalid block dimensions: (\"+rlen+\", \"+clen+\").\");\n+\n+ //2. Correct array lengths\n+ if( size() < nnz )\n+ throw new RuntimeException(\"Incorrect size: \"+size()+\" (expected: \"+nnz+\").\");\n+\n+ //3. Sorted column indices per row\n+ for( int i=0; i<rlen; i++ ) {\n+ if( isEmpty(i) ) continue;\n+ int apos = pos(i);\n+ int alen = size(i);\n+ int[] aix = indexes(i);\n+ double[] avals = values(i);\n+ for (int k = apos + 1; k < apos + alen; k++) {\n+ if (aix[k-1] >= aix[k])\n+ throw new RuntimeException(\"Wrong sparse row ordering, at row: \"\n+ + k + \"with \" + aix[k-1] + \">=\" + aix[k]);\n+ if (avals[k] == 0)\n+ throw new RuntimeException(\"The values are expected to be non zeros \"\n+ + \"but zero at row: \"+ i + \", col pos: \" + k);\n+ }\n+ }\n+\n+ //3. A capacity that is no larger than nnz times resize factor\n+ for( int i=0; i<rlen; i++ )\n+ if( !isEmpty(i) && values(i).length > nnz*RESIZE_FACTOR1 )\n+ throw new RuntimeException(\"The capacity is larger than nnz times a resize factor(=2). \"\n+ + \"Actual length = \" + values(i).length+\", should not exceed \"+nnz*RESIZE_FACTOR1);\n+\nreturn true;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/nn/test/grad_check.dml", "new_path": "scripts/nn/test/grad_check.dml", "diff": "@@ -57,6 +57,7 @@ source(\"nn/test/conv2d_simple.dml\") as conv2d_simple\nsource(\"nn/test/max_pool2d_simple.dml\") as max_pool2d_simple\nsource(\"nn/test/util.dml\") as test_util\nsource(\"nn/util.dml\") as util\n+source(\"nn/layers/elu.dml\") as elu\naffine = function() {\n/*\n@@ -2458,3 +2459,42 @@ two_layer_affine_l2_net_backward = function(matrix[double] X, matrix[double] y,\n[dX, dW1, db1] = affine::backward(dhout, X, W1, b1)\n}\n+elu = function() {\n+ /*\n+ * Gradient check for ELU nonlinearity\n+ * layer.\n+ */\n+ print(\"Grad checking ELU nonlinearity layer with L2 loss.\")\n+ # Generate data\n+ N = 3 # num examples\n+ M = 10 # num neurons\n+\n+ X = rand(rows=N, cols=M)\n+ y = rand(rows=N, cols=M)\n+\n+ out = elu::forward(X, 1)\n+ dout = l2_loss::backward(out, y)\n+ dX = elu::backward(dout, X, 1)\n+\n+ # Grad check\n+ h = 1e-6\n+ print(\" - Grad checking X.\")\n+ for (i in 1:nrow(X)) {\n+ for (j in 1:ncol(X)) {\n+ # Compute numerical derivative\n+ old = as.scalar(X[i,j])\n+ X[i,j] = old - h\n+ outmh = elu::forward(X, 1)\n+ lossmh = l2_loss::forward(outmh, y)\n+ X[i,j] = old + h\n+ outph = elu::forward(X, 1)\n+ lossph = l2_loss::forward(outph, y)\n+ X[i,j] = old # reset\n+ dX_num = (lossph-lossmh) / (2*h) # numerical derivative\n+\n+ # Check error\n+ rel_error = test_util::check_rel_grad_error(as.scalar(dX[i,j]), dX_num, lossph, lossmh)\n+ }\n+ }\n+}\n+\n" }, { "change_type": "MODIFY", "old_path": "scripts/nn/test/run_tests.dml", "new_path": "scripts/nn/test/run_tests.dml", "diff": "@@ -51,6 +51,7 @@ grad_check::conv2d_depthwise()\ngrad_check::conv2d_transpose()\ngrad_check::conv2d_transpose_depthwise()\ngrad_check::dropout()\n+grad_check::elu()\ngrad_check::fm()\ngrad_check::lstm()\ngrad_check::max_pool2d()\n@@ -99,6 +100,7 @@ test::conv2d_transpose()\ntest::conv2d_transpose_depthwise()\ntest::cross_entropy_loss()\ntest::cross_entropy_loss2d()\n+test::elu()\ntest::im2col()\ntest::max_pool2d()\ntest::padding()\n" }, { "change_type": "MODIFY", "old_path": "scripts/nn/test/test.dml", "new_path": "scripts/nn/test/test.dml", "diff": "@@ -40,6 +40,8 @@ source(\"nn/test/max_pool2d_simple.dml\") as max_pool2d_simple\nsource(\"nn/test/util.dml\") as test_util\nsource(\"nn/util.dml\") as util\nsource(\"nn/layers/sigmoid.dml\") as sigmoid\n+source(\"nn/layers/elu.dml\") as elu\n+\nbatch_norm1d = function() {\n/*\n@@ -1093,3 +1095,34 @@ softmax2d = function() {\n}\n}\n+elu = function() {\n+ /*\n+ * Test for ELU function.\n+ */\n+ print(\"Testing ELU function.\")\n+\n+ X = matrix(\"0.3923 -0.2236 -0.3195 -1.2050 1.0445 -0.6332 0.5731 0.5409 -0.3919 -1.0427\", rows = 10, cols = 1)\n+\n+ print(\" - Testing forward\")\n+ out = elu::forward(X, 1)\n+ out_ref = matrix(\"0.3923 -0.2003 -0.2735 -0.7003 1.0445 -0.4691 0.5731 0.5409 -0.3242 -0.6475\", rows = 10, cols = 1)\n+\n+ for (i in 1:nrow(out)) {\n+ for(j in 1:ncol(out)) {\n+ rel_error = test_util::check_rel_error(as.scalar(out[i,j]),\n+ as.scalar(out_ref[i,j]), 1e-3, 1e-3)\n+ }\n+ }\n+\n+ print(\" - Testing backward\")\n+ out = elu::backward(X, X, 1)\n+ out_ref = matrix(\"0.3923 -0.1788 -0.2321 -0.3611 1.0445 -0.3362 0.5731 0.5409 -0.2648 -0.3676\", rows = 10, cols = 1)\n+\n+ for (i in 1:nrow(out)) {\n+ for(j in 1:ncol(out)) {\n+ rel_error = test_util::check_rel_error(as.scalar(out[i,j]),\n+ as.scalar(out_ref[i,j]), 1e-3, 1e-3)\n+ }\n+ }\n+}\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/nn/test/grad_check.dml", "new_path": "scripts/nn/test/grad_check.dml", "diff": "@@ -2469,7 +2469,7 @@ elu = function() {\nN = 3 # num examples\nM = 10 # num neurons\n- X = rand(rows=N, cols=M)\n+ X = rand(rows=N, cols=M, min=-5, max=5)\ny = rand(rows=N, cols=M)\nout = elu::forward(X, 1)\n@@ -2477,7 +2477,7 @@ elu = function() {\ndX = elu::backward(dout, X, 1)\n# Grad check\n- h = 1e-6\n+ h = 1e-5\nprint(\" - Grad checking X.\")\nfor (i in 1:nrow(X)) {\nfor (j in 1:ncol(X)) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/Hop.java", "new_path": "src/main/java/org/apache/sysml/hops/Hop.java", "diff": "@@ -1099,7 +1099,7 @@ public abstract class Hop implements ParseInfo\n// Operations that require a variable number of operands\npublic enum OpOpN {\n- PRINTF, CBIND, RBIND,\n+ PRINTF, CBIND, RBIND, EVAL\n}\npublic enum AggOp {\n@@ -1383,6 +1383,7 @@ public abstract class Hop implements ParseInfo\nHopsOpOpNLops.put(OpOpN.PRINTF, Nary.OperationType.PRINTF);\nHopsOpOpNLops.put(OpOpN.CBIND, Nary.OperationType.CBIND);\nHopsOpOpNLops.put(OpOpN.RBIND, Nary.OperationType.RBIND);\n+ HopsOpOpNLops.put(OpOpN.EVAL, Nary.OperationType.EVAL);\n}\nprotected static final HashMap<Hop.OpOp1, String> HopsOpOp12String;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/NaryOp.java", "new_path": "src/main/java/org/apache/sysml/hops/NaryOp.java", "diff": "@@ -159,7 +159,7 @@ public class NaryOp extends Hop {\nsetRequiresRecompileIfNecessary();\n//ensure cp exec type for single-node operations\n- if ( _op == OpOpN.PRINTF )\n+ if ( _op == OpOpN.PRINTF || _op == OpOpN.EVAL)\n_etype = ExecType.CP;\nreturn _etype;\n@@ -187,6 +187,7 @@ public class NaryOp extends Hop {\nsetDim2(HopRewriteUtils.getMaxInputDim(this, false));\nbreak;\ncase PRINTF:\n+ case EVAL:\n//do nothing:\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/ipa/InterProceduralAnalysis.java", "new_path": "src/main/java/org/apache/sysml/hops/ipa/InterProceduralAnalysis.java", "diff": "@@ -89,7 +89,7 @@ public class InterProceduralAnalysis\nprotected static final boolean INTRA_PROCEDURAL_ANALYSIS = true; //propagate statistics across statement blocks (main/functions)\nprotected static final boolean PROPAGATE_KNOWN_UDF_STATISTICS = true; //propagate statistics for known external functions\nprotected static final boolean ALLOW_MULTIPLE_FUNCTION_CALLS = true; //propagate consistent statistics from multiple calls\n- protected static final boolean REMOVE_UNUSED_FUNCTIONS = true; //remove unused functions (inlined or never called)\n+ protected static final boolean REMOVE_UNUSED_FUNCTIONS = false; //remove unused functions (inlined or never called)\nprotected static final boolean FLAG_FUNCTION_RECOMPILE_ONCE = true; //flag functions which require recompilation inside a loop for full function recompile\nprotected static final boolean REMOVE_UNNECESSARY_CHECKPOINTS = true; //remove unnecessary checkpoints (unconditionally overwritten intermediates)\nprotected static final boolean REMOVE_CONSTANT_BINARY_OPS = true; //remove constant binary operations (e.g., X*ones, where ones=matrix(1,...))\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/lops/Nary.java", "new_path": "src/main/java/org/apache/sysml/lops/Nary.java", "diff": "@@ -32,7 +32,7 @@ import org.apache.sysml.parser.Expression.ValueType;\npublic class Nary extends Lop {\npublic enum OperationType {\n- PRINTF, CBIND, RBIND,\n+ PRINTF, CBIND, RBIND, EVAL\n}\nprivate OperationType operationType;\n@@ -119,6 +119,7 @@ public class Nary extends Lop {\ncase PRINTF:\ncase CBIND:\ncase RBIND:\n+ case EVAL:\nreturn operationType.name().toLowerCase();\ndefault:\nthrow new UnsupportedOperationException(\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "diff": "@@ -24,6 +24,7 @@ import java.util.HashMap;\nimport java.util.HashSet;\nimport org.antlr.v4.runtime.ParserRuleContext;\n+import org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.parser.LanguageException.LanguageErrorCodes;\nimport org.apache.sysml.runtime.util.ConvolutionUtils;\nimport org.apache.sysml.runtime.util.UtilFunctions;\n@@ -379,6 +380,15 @@ public class BuiltinFunctionExpression extends DataIdentifier\nthis.setOutput(output);\nswitch (this.getOpCode()) {\n+ case EVAL:\n+ if (_args.length == 0)\n+ raiseValidateError(\"Function eval should provide at least one argument, i.e., the function name.\", false);\n+ checkValueTypeParam(_args[0], ValueType.STRING);\n+ output.setDataType(DataType.MATRIX);\n+ output.setValueType(ValueType.DOUBLE);\n+ output.setBlockDimensions(ConfigurationManager.getBlocksize(),\n+ ConfigurationManager.getBlocksize());\n+ break;\ncase COLSUM:\ncase COLMAX:\ncase COLMIN:\n@@ -1792,6 +1802,9 @@ public class BuiltinFunctionExpression extends DataIdentifier\nbifop = Expression.BuiltinFunctionOp.BITWSHIFTR;\nelse if ( functionName.equals(\"ifelse\") )\nbifop = Expression.BuiltinFunctionOp.IFELSE;\n+ else if (functionName.equals(\"eval\")) {\n+ bifop = Expression.BuiltinFunctionOp.EVAL;\n+ }\nelse\nreturn null;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -2381,6 +2381,10 @@ public class DMLTranslator\n// Construct the hop based on the type of Builtin function\nswitch (source.getOpCode()) {\n+ case EVAL:\n+ currBuiltinOp = new NaryOp(target.getName(), target.getDataType(), target.getValueType(), OpOpN.EVAL, processAllExpressions(source.getAllExpr(), hops));\n+ break;\n+\ncase COLSUM:\ncurrBuiltinOp = new AggUnaryOp(target.getName(), target.getDataType(), target.getValueType(), AggOp.SUM,\nDirection.Col, expr);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/Expression.java", "new_path": "src/main/java/org/apache/sysml/parser/Expression.java", "diff": "@@ -86,6 +86,7 @@ public abstract class Expression implements ParseInfo\nCUMSUM,\nDIAG,\nEIGEN,\n+ EVAL,\nCONV2D, CONV2D_BACKWARD_FILTER, CONV2D_BACKWARD_DATA, BIAS_ADD, BIAS_MULTIPLY,\nMAX_POOL, AVG_POOL, MAX_POOL_BACKWARD, AVG_POOL_BACKWARD,\nEXP,\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "diff": "@@ -519,7 +519,6 @@ public class DmlSyntacticValidator extends CommonSyntacticValidator implements D\nString functionName = names[1];\nArrayList<ParameterExpression> paramExpression = getParameterExpressionList(ctx.paramExprs);\n-\ncastAsScalarDeprecationCheck(functionName, ctx);\nConvertedDMLSyntax convertedSyntax = convertToDMLSyntax(ctx, namespace, functionName, paramExpression, ctx.name);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/functionobjects/Builtin.java", "new_path": "src/main/java/org/apache/sysml/runtime/functionobjects/Builtin.java", "diff": "@@ -51,7 +51,7 @@ public class Builtin extends ValueFunction\npublic enum BuiltinCode { SIN, COS, TAN, SINH, COSH, TANH, ASIN, ACOS, ATAN, LOG, LOG_NZ, MIN,\nMAX, ABS, SIGN, SQRT, EXP, PLOGP, PRINT, PRINTF, NROW, NCOL, LENGTH, ROUND, MAXINDEX, MININDEX,\n- STOP, CEIL, FLOOR, CUMSUM, CUMPROD, CUMMIN, CUMMAX, INVERSE, SPROP, SIGMOID }\n+ STOP, CEIL, FLOOR, CUMSUM, CUMPROD, CUMMIN, CUMMAX, INVERSE, SPROP, SIGMOID, EVAL }\npublic BuiltinCode bFunc;\nprivate static final boolean FASTMATH = true;\n@@ -81,6 +81,7 @@ public class Builtin extends ValueFunction\nString2BuiltinCode.put( \"plogp\" , BuiltinCode.PLOGP);\nString2BuiltinCode.put( \"print\" , BuiltinCode.PRINT);\nString2BuiltinCode.put( \"printf\" , BuiltinCode.PRINTF);\n+ String2BuiltinCode.put( \"eval\" , BuiltinCode.EVAL);\nString2BuiltinCode.put( \"nrow\" , BuiltinCode.NROW);\nString2BuiltinCode.put( \"ncol\" , BuiltinCode.NCOL);\nString2BuiltinCode.put( \"length\" , BuiltinCode.LENGTH);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/CPInstructionParser.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/CPInstructionParser.java", "diff": "@@ -182,6 +182,7 @@ public class CPInstructionParser extends InstructionParser\nString2CPInstructionType.put( \"printf\" , CPType.BuiltinNary);\nString2CPInstructionType.put( \"cbind\" , CPType.BuiltinNary);\nString2CPInstructionType.put( \"rbind\" , CPType.BuiltinNary);\n+ String2CPInstructionType.put( \"eval\" , CPType.BuiltinNary);\n// Parameterized Builtin Functions\nString2CPInstructionType.put( \"cdf\" , CPType.ParameterizedBuiltin);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/BuiltinNaryCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/BuiltinNaryCPInstruction.java", "diff": "@@ -67,6 +67,9 @@ public abstract class BuiltinNaryCPInstruction extends CPInstruction\nreturn new MatrixBuiltinNaryCPInstruction(null,\nopcode, str, outputOperand, inputOperands);\n}\n+ else if (Nary.OperationType.EVAL.name().equalsIgnoreCase(opcode)) {\n+ return new EvalNaryCPInstruction(null, opcode, str, outputOperand, inputOperands);\n+ }\nthrow new DMLRuntimeException(\"Opcode (\" + opcode + \") not recognized in BuiltinMultipleCPInstruction\");\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/EvalNaryCPInstruction.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.runtime.instructions.cp;\n+\n+import org.apache.sysml.runtime.DMLRuntimeException;\n+import org.apache.sysml.runtime.controlprogram.Program;\n+import org.apache.sysml.runtime.controlprogram.caching.FrameObject;\n+import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\n+import org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+import org.apache.sysml.runtime.matrix.operators.Operator;\n+import org.apache.sysml.runtime.util.DataConverter;\n+\n+import java.util.ArrayList;\n+import java.util.Arrays;\n+\n+/**\n+ * Eval built-in function instruction\n+ * Note: it supports only single matrix[double] output\n+ */\n+public class EvalNaryCPInstruction extends BuiltinNaryCPInstruction {\n+\n+ public EvalNaryCPInstruction(Operator op, String opcode, String istr, CPOperand output, CPOperand... inputs) {\n+ super(op, opcode, istr, output, inputs);\n+ }\n+\n+ @Override\n+ public void processInstruction(ExecutionContext ec) throws DMLRuntimeException {\n+ //1. get the namespace and func\n+ String funcName = ec.getScalarInput(inputs[0]).getStringValue();\n+ if( funcName.contains(Program.KEY_DELIM) )\n+ throw new DMLRuntimeException(\"Eval calls to '\"+funcName+\"', i.e., a function outside \"\n+ + \"the default \"+ \"namespace, are not supported yet. Please call the function directly.\");\n+\n+ // bound the inputs to avoiding being deleted after the function call\n+ CPOperand[] boundInputs = Arrays.copyOfRange(inputs, 1, inputs.length);\n+ ArrayList<String> boundOutputNames = new ArrayList<>();\n+ boundOutputNames.add(output.getName());\n+ ArrayList<String> boundInputNames = new ArrayList<>();\n+ for (CPOperand input : boundInputs) {\n+ boundInputNames.add(input.getName());\n+ }\n+\n+ //2. copy the created output matrix\n+ MatrixObject outputMO = new MatrixObject(ec.getMatrixObject(output.getName()));\n+\n+ //3. call the function\n+ FunctionCallCPInstruction fcpi = new FunctionCallCPInstruction(\n+ null, funcName, boundInputs, boundInputNames, boundOutputNames, \"eval func\");\n+ fcpi.processInstruction(ec);\n+\n+ //4. convert the result to matrix\n+ Data newOutput = ec.getVariable(output);\n+ if (newOutput instanceof MatrixObject) {\n+ return;\n+ }\n+ MatrixBlock mb = null;\n+ if (newOutput instanceof ScalarObject) {\n+ //convert scalar to matrix\n+ mb = new MatrixBlock(((ScalarObject) newOutput).getDoubleValue());\n+ } else if (newOutput instanceof FrameObject) {\n+ //convert frame to matrix\n+ mb = DataConverter.convertToMatrixBlock(((FrameObject) newOutput).acquireRead());\n+ ec.cleanupCacheableData((FrameObject) newOutput);\n+ }\n+ outputMO.acquireModify(mb);\n+ outputMO.release();\n+ ec.setVariable(output.getName(), outputMO);\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/FunctionCallCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/FunctionCallCPInstruction.java", "diff": "@@ -45,7 +45,7 @@ public class FunctionCallCPInstruction extends CPInstruction {\nprivate final ArrayList<String> _boundInputNames;\nprivate final ArrayList<String> _boundOutputNames;\n- private FunctionCallCPInstruction(String namespace, String functName, CPOperand[] boundInputs,\n+ public FunctionCallCPInstruction(String namespace, String functName, CPOperand[] boundInputs,\nArrayList<String> boundInputNames, ArrayList<String> boundOutputNames, String istr) {\nsuper(CPType.External, null, functName, istr);\n_functionName = functName;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "diff": "@@ -88,6 +88,14 @@ import scala.collection.Seq;\npublic class MLContextTest extends MLContextTestBase {\n+ @Test\n+ public void testCreateDMLScriptBasedOnFileAndExecuteEvalTest() {\n+ System.out.println(\"MLContextTest - create DML script based on file and execute\");\n+ setExpectedStdOut(\"10\");\n+ Script script = dmlFromFile(baseDirectory + File.separator + \"eval-test.dml\");\n+ ml.execute(script);\n+ }\n+\n@Test\npublic void testCreateDMLScriptBasedOnStringAndExecute() {\nSystem.out.println(\"MLContextTest - create DML script based on string and execute\");\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/org/apache/sysml/api/mlcontext/eval-test.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+# product between matrix and scalar\n+f1 = function (matrix[double] M, double factor) return (double res) {\n+ res = prod(M) * factor\n+}\n+\n+# replace the matrix with a scalar and then use calculate the product\n+f2 = function (matrix[double] M, double r) return (double res) {\n+ R = replace(target=M, pattern=1, replacement=r)\n+ res = f1(R, 10)\n+}\n+\n+# production of two matrix\n+f3 = function (matrix[double] M1, matrix[double] M2) return (matrix[double] res) {\n+ res = M1 %*% M2\n+}\n+\n+f4 = function (matrix[double] M1, matrix[double] M2) return (matrix[double] res) {\n+ res = M1 %*% M2\n+}\n+\n+# some variables\n+X = matrix(\"1 2 3 4\", rows=2, cols=2)\n+y = 10\n+\n+R1 = eval(\"f1\", X, y)\n+R2 = eval(\"f2\", X, y)\n+R3 = eval(\"f3\", X, X)\n+for(i in 3:4)\n+ R4 = eval(\"f\"+i, X, X);\n+\n+print(toString(R1))\n+print(toString(R2))\n+print(toString(R3))\n+print(toString(R4))\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<properties>\n<hadoop.version>2.6.0</hadoop.version>\n- <antlr.version>4.5.3</antlr.version>\n+ <antlr.version>4.7.1</antlr.version>\n<spark.version>2.1.0</spark.version>\n<scala.version>2.11.8</scala.version>\n<scala.binary.version>2.11</scala.binary.version>\n<include>**/integration/functions/data/*Suite.java</include>\n<include>**/integration/functions/gdfo/*Suite.java</include>\n<include>**/integration/functions/sparse/*Suite.java</include>\n+ <include>**/integration/functions/codegenalg/*Suite.java</include>\n<include>**/integration/functions/**/*Test*.java</include>\n<include>**/integration/mlcontext/*Suite.java</include>\n<include>**/integration/mlcontext/algorithms/*Suite.java</include>\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<exclude>**/*.ipynb</exclude>\n<!-- Generated antlr files -->\n<exclude>src/main/java/*.tokens</exclude>\n+ <exclude>**/*.interp</exclude>\n<!-- Generated python files -->\n<exclude>src/main/python/systemml.egg-info/**</exclude>\n<!-- Sphinx reStructuredText files -->\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "diff": "@@ -690,6 +690,10 @@ public class SpoofCompiler\nCNodeTpl tpl = e.getValue().getValue();\nHop[] inHops = e.getValue().getKey();\n+ //remove invalid plans with null inputs\n+ if( Arrays.stream(inHops).anyMatch(h -> (h==null)) )\n+ continue;\n+\n//perform simplifications and cse rewrites\ntpl = rewriter.simplifyCPlan(tpl);\ntpl = cse.eliminateCommonSubexpressions(tpl);\n@@ -697,7 +701,7 @@ public class SpoofCompiler\n//update input hops (order-preserving)\nHashSet<Long> inputHopIDs = tpl.getInputHopIDs(false);\ninHops = Arrays.stream(inHops)\n- .filter(p -> inputHopIDs.contains(p.getHopID()))\n+ .filter(p -> p != null && inputHopIDs.contains(p.getHopID()))\n.toArray(Hop[]::new);\ncplans2.put(e.getKey(), new Pair<>(inHops, tpl));\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/dml/DMLParserWrapper.java", "new_path": "src/main/java/org/apache/sysml/parser/dml/DMLParserWrapper.java", "diff": "@@ -71,6 +71,7 @@ import org.apache.sysml.parser.dml.DmlParser.StatementContext;\n* If in future we intend to make it multi-threaded, look at cleanUpState method and resolve the dependency accordingly.\n*\n*/\n+@SuppressWarnings(\"deprecation\")\npublic class DMLParserWrapper extends ParserWrapper\n{\nprivate static final Log LOG = LogFactory.getLog(DMLScript.class.getName());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/pydml/PyDMLParserWrapper.java", "new_path": "src/main/java/org/apache/sysml/parser/pydml/PyDMLParserWrapper.java", "diff": "@@ -54,6 +54,7 @@ import org.apache.sysml.parser.pydml.PydmlParser.StatementContext;\n* Note: ExpressionInfo and StatementInfo are simply wrapper objects and are reused in both DML and PyDML parsers.\n*\n*/\n+@SuppressWarnings(\"deprecation\")\npublic class PyDMLParserWrapper extends ParserWrapper\n{\nprivate static final Log LOG = LogFactory.getLog(DMLScript.class.getName());\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteSplitDagDataDependentOperators.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteSplitDagDataDependentOperators.java", "diff": "@@ -21,6 +21,7 @@ package org.apache.sysml.hops.rewrite;\nimport java.util.ArrayList;\nimport java.util.Arrays;\n+import java.util.HashMap;\nimport java.util.HashSet;\nimport java.util.List;\n@@ -112,8 +113,8 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\n//unless there are transient reads w/ the same variable name in the current dag which can\n//lead to invalid reordering if variable consumers are not feeding into the candidate op.\nboolean hasTWrites = hasTransientWriteParents(c);\n- boolean moveTWrite = hasTWrites ? HopRewriteUtils.rHasSimpleReadChain(c,\n- getFirstTransientWriteParent(c).getName()) : false;\n+ boolean moveTWrite = hasTWrites ? HopRewriteUtils.rHasSimpleReadChain(\n+ c, getFirstTransientWriteParent(c).getName()) : false;\nString varname = null;\nlong rlen = c.getDim1();\n@@ -170,8 +171,8 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\n}\n//add data-dependent operator sub dag to first statement block\n- DataOp twrite = new DataOp(varname, c.getDataType(), c.getValueType(),\n- c, DataOpTypes.TRANSIENTWRITE, null);\n+ DataOp twrite = new DataOp(varname, c.getDataType(),\n+ c.getValueType(), c, DataOpTypes.TRANSIENTWRITE, null);\ntwrite.setVisited();\ntwrite.setOutputParams(rlen, clen, nnz, update, brlen, bclen);\nHopRewriteUtils.copyLineNumbers(c, twrite);\n@@ -349,29 +350,27 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\nfor( Hop h : rootsSB2 )\nrProbeAndAddHopsToCandidateSet(h, probeSet, candSet);\n- //step 3: create additional cuts\n- for( Pair<Hop,Hop> p : candSet )\n- {\n- String varname = createCutVarName(false);\n-\n+ //step 3: create additional cuts with reuse for common references\n+ HashMap<Long, DataOp> reuseTRead = new HashMap<>();\n+ for( Pair<Hop,Hop> p : candSet ) {\nHop hop = p.getKey();\nHop c = p.getValue();\n- DataOp tread = new DataOp(varname, c.getDataType(), c.getValueType(), DataOpTypes.TRANSIENTREAD,\n- null, c.getDim1(), c.getDim2(), c.getNnz(), c.getUpdateType(), c.getRowsInBlock(), c.getColsInBlock());\n+ DataOp tread = reuseTRead.get(c.getHopID());\n+ if( tread == null ) {\n+ String varname = createCutVarName(false);\n+\n+ tread = new DataOp(varname, c.getDataType(), c.getValueType(), DataOpTypes.TRANSIENTREAD, null,\n+ c.getDim1(), c.getDim2(), c.getNnz(), c.getUpdateType(), c.getRowsInBlock(), c.getColsInBlock());\ntread.setVisited();\nHopRewriteUtils.copyLineNumbers(c, tread);\n+ reuseTRead.put(c.getHopID(), tread);\nDataOp twrite = new DataOp(varname, c.getDataType(), c.getValueType(), c, DataOpTypes.TRANSIENTWRITE, null);\ntwrite.setVisited();\ntwrite.setOutputParams(c.getDim1(), c.getDim2(), c.getNnz(), c.getUpdateType(), c.getRowsInBlock(), c.getColsInBlock());\nHopRewriteUtils.copyLineNumbers(c, twrite);\n- //create additional cut by rewriting both hop dags\n- int pos = HopRewriteUtils.getChildReferencePos(hop, c);\n- HopRewriteUtils.removeChildReferenceByPos(hop, c, pos);\n- HopRewriteUtils.addChildReference(hop, tread, pos);\n-\n//update live in and out of new statement block (for piggybacking)\nDataIdentifier diVar = new DataIdentifier(varname);\ndiVar.setDimensions(c.getDim1(), c.getDim2());\n@@ -383,6 +382,12 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\nrootsSB1.add(twrite);\n}\n+\n+ //create additional cut by rewriting both hop dags\n+ int pos = HopRewriteUtils.getChildReferencePos(hop, c);\n+ HopRewriteUtils.removeChildReferenceByPos(hop, c, pos);\n+ HopRewriteUtils.addChildReference(hop, tread, pos);\n+ }\n}\nprivate void rAddHopsToProbeSet( Hop hop, HashSet<Hop> probeSet )\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/recompile/LiteralReplaceCastScalarReadTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/recompile/LiteralReplaceCastScalarReadTest.java", "diff": "@@ -21,7 +21,7 @@ package org.apache.sysml.test.integration.functions.recompile;\nimport org.junit.Assert;\nimport org.junit.Test;\n-\n+import org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.lops.UnaryCP;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.test.integration.AutomatedTestBase;\n@@ -29,51 +29,42 @@ import org.apache.sysml.test.integration.TestConfiguration;\nimport org.apache.sysml.test.utils.TestUtils;\nimport org.apache.sysml.utils.Statistics;\n-/**\n- *\n- */\npublic class LiteralReplaceCastScalarReadTest extends AutomatedTestBase\n{\n-\nprivate final static String TEST_NAME = \"LiteralReplaceCastScalar\";\nprivate final static String TEST_DIR = \"functions/recompile/\";\nprivate final static String TEST_CLASS_DIR = TEST_DIR +\nLiteralReplaceCastScalarReadTest.class.getSimpleName() + \"/\";\n@Override\n- public void setUp()\n- {\n+ public void setUp() {\nTestUtils.clearAssertionInformation();\naddTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[] { \"R\" }));\n}\n@Test\n- public void testRemoveCastsInputInteger()\n- {\n+ public void testRemoveCastsInputInteger() {\nrunScalarCastTest(ValueType.INT);\n}\n@Test\n- public void testRemoveCastsInputDouble()\n- {\n+ public void testRemoveCastsInputDouble() {\nrunScalarCastTest(ValueType.DOUBLE);\n}\n@Test\n- public void testRemoveCastsInputBoolean()\n- {\n+ public void testRemoveCastsInputBoolean() {\nrunScalarCastTest(ValueType.BOOLEAN);\n}\n+ private void runScalarCastTest( ValueType vt ) {\n+ boolean oldCF = OptimizerUtils.ALLOW_CONSTANT_FOLDING;\n- /**\n- *\n- * @param vt\n- */\n- private void runScalarCastTest( ValueType vt )\n- {\n+ try {\nTestConfiguration config = getTestConfiguration(TEST_NAME);\n+ loadTestConfiguration(config);\n+ OptimizerUtils.ALLOW_CONSTANT_FOLDING = false;\n// input value\nString val = null;\n@@ -90,8 +81,6 @@ public class LiteralReplaceCastScalarReadTest extends AutomatedTestBase\n//note: stats required for runtime check of rewrite\nprogramArgs = new String[]{\"-explain\",\"-stats\",\"-args\", val };\n- loadTestConfiguration(config);\n-\nrunTest(true, false, null, -1);\n//CHECK cast replacement and sum replacement\n@@ -100,5 +89,8 @@ public class LiteralReplaceCastScalarReadTest extends AutomatedTestBase\nAssert.assertEquals(false, Statistics.getCPHeavyHitterOpCodes().contains(UnaryCP.CAST_AS_BOOLEAN_OPCODE));\nAssert.assertEquals(false, Statistics.getCPHeavyHitterOpCodes().contains(\"uak+\")); //sum\n}\n-\n+ finally {\n+ OptimizerUtils.ALLOW_CONSTANT_FOLDING = oldCF;\n+ }\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/assembly/bin/NOTICE", "new_path": "src/assembly/bin/NOTICE", "diff": "Apache SystemML\n-Copyright [2015-2017] The Apache Software Foundation\n+Copyright [2015-2018] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/extra/NOTICE", "new_path": "src/assembly/extra/NOTICE", "diff": "Apache SystemML\n-Copyright [2015-2017] The Apache Software Foundation\n+Copyright [2015-2018] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/jar/NOTICE", "new_path": "src/assembly/jar/NOTICE", "diff": "Apache SystemML\n-Copyright [2015-2017] The Apache Software Foundation\n+Copyright [2015-2018] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/lite/NOTICE", "new_path": "src/assembly/lite/NOTICE", "diff": "Apache SystemML\n-Copyright [2015-2017] The Apache Software Foundation\n+Copyright [2015-2018] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/standalone-jar/NOTICE", "new_path": "src/assembly/standalone-jar/NOTICE", "diff": "Apache SystemML\n-Copyright [2015-2017] The Apache Software Foundation\n+Copyright [2015-2018] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "diff": "@@ -380,6 +380,7 @@ public class SparkExecutionContext extends ExecutionContext\n//keep rdd handle for future operations on it\nRDDObject rddhandle = new RDDObject(rdd);\nrddhandle.setHDFSFile(fromFile);\n+ rddhandle.setParallelizedRDD(!fromFile);\nmo.setRDDHandle(rddhandle);\n}\n//CASE 3: non-dirty (file exists on HDFS)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "diff": "package org.apache.sysml.runtime.controlprogram.context;\nimport java.io.IOException;\n+import java.util.Arrays;\nimport java.util.HashMap;\nimport java.util.LinkedList;\nimport java.util.List;\n+import java.util.stream.Collectors;\n+import java.util.stream.LongStream;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n@@ -684,37 +687,17 @@ public class SparkExecutionContext extends ExecutionContext\nthrows DMLRuntimeException\n{\nlong t0 = DMLScript.STATISTICS ? System.nanoTime() : 0;\n- LinkedList<Tuple2<MatrixIndexes,MatrixBlock>> list = new LinkedList<>();\n+ List<Tuple2<MatrixIndexes,MatrixBlock>> list = null;\n- if( src.getNumRows() <= brlen\n- && src.getNumColumns() <= bclen )\n- {\n- list.addLast(new Tuple2<>(new MatrixIndexes(1,1), src));\n- }\n- else\n- {\n- boolean sparse = src.isInSparseFormat();\n-\n- //create and write subblocks of matrix\n- for(int blockRow = 0; blockRow < (int)Math.ceil(src.getNumRows()/(double)brlen); blockRow++)\n- for(int blockCol = 0; blockCol < (int)Math.ceil(src.getNumColumns()/(double)bclen); blockCol++)\n- {\n- int maxRow = (blockRow*brlen + brlen < src.getNumRows()) ? brlen : src.getNumRows() - blockRow*brlen;\n- int maxCol = (blockCol*bclen + bclen < src.getNumColumns()) ? bclen : src.getNumColumns() - blockCol*bclen;\n-\n- MatrixBlock block = new MatrixBlock(maxRow, maxCol, sparse);\n-\n- int row_offset = blockRow*brlen;\n- int col_offset = blockCol*bclen;\n-\n- //copy submatrix to block\n- src.slice( row_offset, row_offset+maxRow-1,\n- col_offset, col_offset+maxCol-1, block );\n-\n- //append block to sequence file\n- MatrixIndexes indexes = new MatrixIndexes(blockRow+1, blockCol+1);\n- list.addLast(new Tuple2<>(indexes, block));\n+ if( src.getNumRows() <= brlen && src.getNumColumns() <= bclen ) {\n+ list = Arrays.asList(new Tuple2<>(new MatrixIndexes(1,1), src));\n}\n+ else {\n+ MatrixCharacteristics mc = new MatrixCharacteristics(\n+ src.getNumRows(), src.getNumColumns(), brlen, bclen, src.getNonZeros());\n+ list = LongStream.range(0, mc.getNumBlocks()).parallel()\n+ .mapToObj(i -> createIndexedBlock(src, mc, i))\n+ .collect(Collectors.toList());\n}\nJavaPairRDD<MatrixIndexes,MatrixBlock> result = sc.parallelizePairs(list);\n@@ -726,6 +709,28 @@ public class SparkExecutionContext extends ExecutionContext\nreturn result;\n}\n+ private static Tuple2<MatrixIndexes,MatrixBlock> createIndexedBlock(MatrixBlock mb, MatrixCharacteristics mc, long ix) {\n+ try {\n+ //compute block indexes\n+ long blockRow = ix / mc.getNumColBlocks();\n+ long blockCol = ix % mc.getNumColBlocks();\n+ //compute block sizes\n+ int maxRow = UtilFunctions.computeBlockSize(mc.getRows(), blockRow+1, mc.getRowsPerBlock());\n+ int maxCol = UtilFunctions.computeBlockSize(mc.getCols(), blockCol+1, mc.getColsPerBlock());\n+ //copy sub-matrix to block\n+ MatrixBlock block = new MatrixBlock(maxRow, maxCol, mb.isInSparseFormat());\n+ int row_offset = (int)blockRow*mc.getRowsPerBlock();\n+ int col_offset = (int)blockCol*mc.getColsPerBlock();\n+ block = mb.slice( row_offset, row_offset+maxRow-1,\n+ col_offset, col_offset+maxCol-1, block );\n+ //create key-value pair\n+ return new Tuple2<>(new MatrixIndexes(blockRow+1, blockCol+1), block);\n+ }\n+ catch(DMLRuntimeException ex) {\n+ throw new RuntimeException(ex);\n+ }\n+ }\n+\npublic static JavaPairRDD<Long,FrameBlock> toFrameJavaPairRDD(JavaSparkContext sc, FrameBlock src)\nthrows DMLRuntimeException\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<version>18</version>\n</parent>\n<groupId>org.apache.systemml</groupId>\n- <version>1.2.0-SNAPSHOT</version>\n+ <version>1.1.0</version>\n<artifactId>systemml</artifactId>\n<packaging>jar</packaging>\n<name>SystemML</name>\n<connection>scm:git:[email protected]:apache/systemml</connection>\n<developerConnection>scm:git:https://git-wip-us.apache.org/repos/asf/systemml</developerConnection>\n<url>https://git-wip-us.apache.org/repos/asf?p=systemml.git</url>\n- <tag>HEAD</tag>\n+ <tag>v1.1.0-rc2</tag>\n</scm>\n<issueManagement>\n<system>JIRA</system>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptExecutor.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptExecutor.java", "diff": "@@ -93,22 +93,7 @@ import org.apache.sysml.utils.Statistics;\n* </ol>\n* <p>\n* Modifications to these steps can be accomplished by subclassing\n- * ScriptExecutor. For example, the following code will turn off the global data\n- * flow optimization check by subclassing ScriptExecutor and overriding the\n- * globalDataFlowOptimization method.\n- * </p>\n- *\n- * <code>ScriptExecutor scriptExecutor = new ScriptExecutor() {\n- * <br>&nbsp;&nbsp;// turn off global data flow optimization check\n- * <br>&nbsp;&nbsp;@Override\n- * <br>&nbsp;&nbsp;protected void globalDataFlowOptimization() {\n- * <br>&nbsp;&nbsp;&nbsp;&nbsp;return;\n- * <br>&nbsp;&nbsp;}\n- * <br>};\n- * <br>ml.execute(script, scriptExecutor);</code>\n- * <p>\n- *\n- * For more information, please see the {@link #execute} method.\n+ * ScriptExecutor. For more information, please see the {@link #execute} method.\n*/\npublic class ScriptExecutor {\n@@ -307,7 +292,6 @@ public class ScriptExecutor {\n* <li>{@link #constructLops()}</li>\n* <li>{@link #generateRuntimeProgram()}</li>\n* <li>{@link #showExplanation()}</li>\n- * <li>{@link #globalDataFlowOptimization()}</li>\n* <li>{@link #countCompiledMRJobsAndSparkInstructions()}</li>\n* <li>{@link #initializeCachingAndScratchSpace()}</li>\n* <li>{@link #cleanupRuntimeProgram()}</li>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/DMLScript.java", "new_path": "src/main/java/org/apache/sysml/api/DMLScript.java", "diff": "@@ -235,10 +235,9 @@ public class DMLScript\n*\n* @param args command-line arguments\n* @throws IOException if an IOException occurs\n- * @throws DMLException if a DMLException occurs\n*/\npublic static void main(String[] args)\n- throws IOException, DMLException\n+ throws IOException\n{\nConfiguration conf = new Configuration(ConfigurationManager.getCachedJobConf());\nString[] otherArgs = new GenericOptionsParser(conf, args).getRemainingArgs();\n@@ -452,11 +451,8 @@ public class DMLScript\n* @param conf Hadoop configuration\n* @param args arguments\n* @return true if success, false otherwise\n- * @throws DMLException if DMLException occurs\n*/\n- public static boolean executeScript( Configuration conf, String[] args )\n- throws DMLException\n- {\n+ public static boolean executeScript( Configuration conf, String[] args ) {\n//parse arguments and set execution properties\nRUNTIME_PLATFORM oldrtplatform = rtplatform; //keep old rtplatform\nExplainType oldexplain = EXPLAIN; //keep old explain\n@@ -956,9 +952,7 @@ public class DMLScript\nreturn dateFormat.format(date);\n}\n- private static void cleanSystemMLWorkspace()\n- throws DMLException\n- {\n+ private static void cleanSystemMLWorkspace() {\ntry\n{\n//read the default config\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "diff": "@@ -186,11 +186,8 @@ public class Connection implements Closeable\n* @param inputs string array of input variables to register\n* @param outputs string array of output variables to register\n* @return PreparedScript object representing the precompiled script\n- * @throws DMLException if DMLException occurs\n*/\n- public PreparedScript prepareScript( String script, String[] inputs, String[] outputs)\n- throws DMLException\n- {\n+ public PreparedScript prepareScript( String script, String[] inputs, String[] outputs) {\nreturn prepareScript(script, inputs, outputs, false);\n}\n@@ -202,11 +199,8 @@ public class Connection implements Closeable\n* @param outputs string array of output variables to register\n* @param parsePyDML {@code true} if PyDML, {@code false} if DML\n* @return PreparedScript object representing the precompiled script\n- * @throws DMLException if DMLException occurs\n*/\n- public PreparedScript prepareScript( String script, String[] inputs, String[] outputs, boolean parsePyDML)\n- throws DMLException\n- {\n+ public PreparedScript prepareScript( String script, String[] inputs, String[] outputs, boolean parsePyDML) {\nreturn prepareScript(script, new HashMap<String,String>(), inputs, outputs, parsePyDML);\n}\n@@ -219,11 +213,8 @@ public class Connection implements Closeable\n* @param outputs string array of output variables to register\n* @param parsePyDML {@code true} if PyDML, {@code false} if DML\n* @return PreparedScript object representing the precompiled script\n- * @throws DMLException if DMLException occurs\n*/\n- public PreparedScript prepareScript( String script, Map<String, String> args, String[] inputs, String[] outputs, boolean parsePyDML)\n- throws DMLException\n- {\n+ public PreparedScript prepareScript( String script, Map<String, String> args, String[] inputs, String[] outputs, boolean parsePyDML) {\nDMLScript.SCRIPT_TYPE = parsePyDML ? ScriptType.PYDML : ScriptType.DML;\n//check for valid names of passed arguments\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/JMLCUtils.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/JMLCUtils.java", "diff": "@@ -48,21 +48,16 @@ public class JMLCUtils\n* @param prog the DML/PyDML program\n* @param outputs registered output variables\n*/\n- public static void cleanupRuntimeProgram( Program prog, String[] outputs)\n- {\n+ public static void cleanupRuntimeProgram( Program prog, String[] outputs) {\nMap<String, FunctionProgramBlock> funcMap = prog.getFunctionProgramBlocks();\nHashSet<String> blacklist = new HashSet<>(Arrays.asList(outputs));\n-\n- if( funcMap != null && !funcMap.isEmpty() )\n- {\n- for( Entry<String, FunctionProgramBlock> e : funcMap.entrySet() )\n- {\n+ if( funcMap != null && !funcMap.isEmpty() ) {\n+ for( Entry<String, FunctionProgramBlock> e : funcMap.entrySet() ) {\nFunctionProgramBlock fpb = e.getValue();\nfor( ProgramBlock pb : fpb.getChildBlocks() )\nrCleanupRuntimeProgram(pb, blacklist);\n}\n}\n-\nfor( ProgramBlock pb : prog.getProgramBlocks() )\nrCleanupRuntimeProgram(pb, blacklist);\n}\n@@ -73,24 +68,20 @@ public class JMLCUtils\n* @param pb program block\n* @param outputs registered output variables\n*/\n- public static void rCleanupRuntimeProgram( ProgramBlock pb, HashSet<String> outputs )\n- {\n- if( pb instanceof WhileProgramBlock )\n- {\n+ public static void rCleanupRuntimeProgram( ProgramBlock pb, HashSet<String> outputs ) {\n+ if( pb instanceof WhileProgramBlock ) {\nWhileProgramBlock wpb = (WhileProgramBlock)pb;\nfor( ProgramBlock pbc : wpb.getChildBlocks() )\nrCleanupRuntimeProgram(pbc,outputs);\n}\n- else if( pb instanceof IfProgramBlock )\n- {\n+ else if( pb instanceof IfProgramBlock ) {\nIfProgramBlock ipb = (IfProgramBlock)pb;\nfor( ProgramBlock pbc : ipb.getChildBlocksIfBody() )\nrCleanupRuntimeProgram(pbc,outputs);\nfor( ProgramBlock pbc : ipb.getChildBlocksElseBody() )\nrCleanupRuntimeProgram(pbc,outputs);\n}\n- else if( pb instanceof ForProgramBlock )\n- {\n+ else if( pb instanceof ForProgramBlock ) {\nForProgramBlock fpb = (ForProgramBlock)pb;\nfor( ProgramBlock pbc : fpb.getChildBlocks() )\nrCleanupRuntimeProgram(pbc,outputs);\n@@ -121,13 +112,10 @@ public class JMLCUtils\n* @param outputs registered output variables\n* @return list of instructions\n*/\n- public static ArrayList<Instruction> cleanupRuntimeInstructions( ArrayList<Instruction> insts, HashSet<String> outputs )\n- {\n+ public static ArrayList<Instruction> cleanupRuntimeInstructions( ArrayList<Instruction> insts, HashSet<String> outputs ) {\nArrayList<Instruction> ret = new ArrayList<>();\n-\nfor( Instruction inst : insts ) {\n- if( inst instanceof VariableCPInstruction && ((VariableCPInstruction)inst).isRemoveVariable() )\n- {\n+ if( inst instanceof VariableCPInstruction && ((VariableCPInstruction)inst).isRemoveVariable() ) {\nArrayList<String> currRmVar = new ArrayList<>();\nfor( CPOperand input : ((VariableCPInstruction)inst).getInputs() )\nif( !outputs.contains(input.getName()) )\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "diff": "@@ -103,8 +103,7 @@ public class PreparedScript implements ConfigurableAPI\n* @param dmlconf dml configuration\n* @param cconf compiler configuration\n*/\n- protected PreparedScript( Program prog, String[] inputs, String[] outputs, DMLConfig dmlconf, CompilerConfig cconf )\n- {\n+ protected PreparedScript( Program prog, String[] inputs, String[] outputs, DMLConfig dmlconf, CompilerConfig cconf ) {\n_prog = prog;\n_vars = new LocalVariableMap();\n@@ -164,9 +163,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param scalar boolean value\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, boolean scalar) throws DMLException {\n+ public void setScalar(String varname, boolean scalar) {\nsetScalar(varname, scalar, false);\n}\n@@ -176,9 +174,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param scalar boolean value\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, boolean scalar, boolean reuse) throws DMLException {\n+ public void setScalar(String varname, boolean scalar, boolean reuse) {\nsetScalar(varname, new BooleanObject(scalar), reuse);\n}\n@@ -187,9 +184,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param scalar long value\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, long scalar) throws DMLException {\n+ public void setScalar(String varname, long scalar) {\nsetScalar(varname, scalar, false);\n}\n@@ -199,9 +195,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param scalar long value\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, long scalar, boolean reuse) throws DMLException {\n+ public void setScalar(String varname, long scalar, boolean reuse) {\nsetScalar(varname, new IntObject(scalar), reuse);\n}\n@@ -209,9 +204,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param scalar double value\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, double scalar) throws DMLException {\n+ public void setScalar(String varname, double scalar) {\nsetScalar(varname, scalar, false);\n}\n@@ -221,9 +215,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param scalar double value\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, double scalar, boolean reuse) throws DMLException {\n+ public void setScalar(String varname, double scalar, boolean reuse) {\nsetScalar(varname, new DoubleObject(scalar), reuse);\n}\n@@ -232,9 +225,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param scalar string value\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, String scalar) throws DMLException {\n+ public void setScalar(String varname, String scalar) {\nsetScalar(varname, scalar, false);\n}\n@@ -244,9 +236,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param scalar string value\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, String scalar, boolean reuse) throws DMLException {\n+ public void setScalar(String varname, String scalar, boolean reuse) {\nsetScalar(varname, new StringObject(scalar), reuse);\n}\n@@ -258,14 +249,10 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param scalar scalar object\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setScalar(String varname, ScalarObject scalar, boolean reuse)\n- throws DMLException\n- {\n+ public void setScalar(String varname, ScalarObject scalar, boolean reuse) {\nif( !_inVarnames.contains(varname) )\nthrow new DMLException(\"Unspecified input variable: \"+varname);\n-\n_vars.put(varname, scalar);\n}\n@@ -274,9 +261,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param matrix two-dimensional double array matrix representation\n- * @throws DMLException if DMLException occurs\n*/\n- public void setMatrix(String varname, double[][] matrix) throws DMLException {\n+ public void setMatrix(String varname, double[][] matrix) {\nsetMatrix(varname, matrix, false);\n}\n@@ -286,9 +272,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param matrix two-dimensional double array matrix representation\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setMatrix(String varname, double[][] matrix, boolean reuse) throws DMLException {\n+ public void setMatrix(String varname, double[][] matrix, boolean reuse) {\nsetMatrix(varname, DataConverter.convertToMatrixBlock(matrix), reuse);\n}\n@@ -300,11 +285,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param matrix matrix represented as a MatrixBlock\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setMatrix(String varname, MatrixBlock matrix, boolean reuse)\n- throws DMLException\n- {\n+ public void setMatrix(String varname, MatrixBlock matrix, boolean reuse) {\nif( !_inVarnames.contains(varname) )\nthrow new DMLException(\"Unspecified input variable: \"+varname);\n@@ -330,9 +312,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param varname input variable name\n* @param frame two-dimensional string array frame representation\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame) throws DMLException {\n+ public void setFrame(String varname, String[][] frame) {\nsetFrame(varname, frame, false);\n}\n@@ -342,9 +323,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param frame two-dimensional string array frame representation\n* @param schema list representing the types of the frame columns\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame, List<ValueType> schema) throws DMLException {\n+ public void setFrame(String varname, String[][] frame, List<ValueType> schema) {\nsetFrame(varname, frame, schema, false);\n}\n@@ -355,9 +335,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param frame two-dimensional string array frame representation\n* @param schema list representing the types of the frame columns\n* @param colnames frame column names\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame, List<ValueType> schema, List<String> colnames) throws DMLException {\n+ public void setFrame(String varname, String[][] frame, List<ValueType> schema, List<String> colnames) {\nsetFrame(varname, frame, schema, colnames, false);\n}\n@@ -367,9 +346,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param frame two-dimensional string array frame representation\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame, boolean reuse) throws DMLException {\n+ public void setFrame(String varname, String[][] frame, boolean reuse) {\nsetFrame(varname, DataConverter.convertToFrameBlock(frame), reuse);\n}\n@@ -380,9 +358,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param frame two-dimensional string array frame representation\n* @param schema list representing the types of the frame columns\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame, List<ValueType> schema, boolean reuse) throws DMLException {\n+ public void setFrame(String varname, String[][] frame, List<ValueType> schema, boolean reuse) {\nsetFrame(varname, DataConverter.convertToFrameBlock(frame, schema.toArray(new ValueType[0])), reuse);\n}\n@@ -394,9 +371,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param schema list representing the types of the frame columns\n* @param colnames frame column names\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, String[][] frame, List<ValueType> schema, List<String> colnames, boolean reuse) throws DMLException {\n+ public void setFrame(String varname, String[][] frame, List<ValueType> schema, List<String> colnames, boolean reuse) {\nsetFrame(varname, DataConverter.convertToFrameBlock( frame,\nschema.toArray(new ValueType[0]), colnames.toArray(new String[0])), reuse);\n}\n@@ -409,11 +385,8 @@ public class PreparedScript implements ConfigurableAPI\n* @param varname input variable name\n* @param frame frame represented as a FrameBlock\n* @param reuse if {@code true}, preserve value over multiple {@code executeScript} calls\n- * @throws DMLException if DMLException occurs\n*/\n- public void setFrame(String varname, FrameBlock frame, boolean reuse)\n- throws DMLException\n- {\n+ public void setFrame(String varname, FrameBlock frame, boolean reuse) {\nif( !_inVarnames.contains(varname) )\nthrow new DMLException(\"Unspecified input variable: \"+varname);\n@@ -445,11 +418,8 @@ public class PreparedScript implements ConfigurableAPI\n* result variables according to bound and registered outputs.\n*\n* @return ResultVariables object encapsulating output results\n- * @throws DMLException if DMLException occurs\n*/\n- public ResultVariables executeScript()\n- throws DMLException\n- {\n+ public ResultVariables executeScript() {\n//add reused variables\n_vars.putAll(_inVarReuse);\n@@ -484,9 +454,8 @@ public class PreparedScript implements ConfigurableAPI\n* Explain the DML/PyDML program and view result as a string.\n*\n* @return string results of explain\n- * @throws DMLException if DMLException occurs\n*/\n- public String explain() throws DMLException {\n+ public String explain() {\nreturn Explain.explain(_prog);\n}\n@@ -501,11 +470,8 @@ public class PreparedScript implements ConfigurableAPI\n*\n* @param fnamespace function namespace, null for default namespace\n* @param fnames function name\n- * @throws DMLException if any compilation error occurs\n*/\n- public void enableFunctionRecompile(String fnamespace, String... fnames)\n- throws DMLException\n- {\n+ public void enableFunctionRecompile(String fnamespace, String... fnames) {\n//handle default name space\nif( fnamespace == null )\nfnamespace = DMLProgram.DEFAULT_NAMESPACE;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/ResultVariables.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/ResultVariables.java", "diff": "@@ -68,9 +68,8 @@ public class ResultVariables\n*\n* @param varname output variable name\n* @return matrix as a two-dimensional double array\n- * @throws DMLException if DMLException occurs\n*/\n- public double[][] getMatrix(String varname) throws DMLException {\n+ public double[][] getMatrix(String varname) {\nreturn DataConverter.convertToDoubleMatrix(getMatrixBlock(varname));\n}\n@@ -80,9 +79,8 @@ public class ResultVariables\n*\n* @param varname output variable name\n* @return matrix as matrix block\n- * @throws DMLException if DMLException occurs\n*/\n- public MatrixBlock getMatrixBlock(String varname) throws DMLException {\n+ public MatrixBlock getMatrixBlock(String varname) {\nData dat = _out.get(varname);\nif( dat == null )\nthrow new DMLException(\"Non-existent output variable: \"+varname);\n@@ -103,9 +101,8 @@ public class ResultVariables\n*\n* @param varname output variable name\n* @return frame as a two-dimensional string array\n- * @throws DMLException if DMLException occurs\n*/\n- public String[][] getFrame(String varname) throws DMLException {\n+ public String[][] getFrame(String varname) {\nreturn DataConverter.convertToStringFrame(getFrameBlock(varname));\n}\n@@ -115,11 +112,8 @@ public class ResultVariables\n*\n* @param varname output variable name\n* @return frame as a frame block\n- * @throws DMLException if DMLException occurs\n*/\n- public FrameBlock getFrameBlock(String varname)\n- throws DMLException\n- {\n+ public FrameBlock getFrameBlock(String varname) {\nData dat = _out.get(varname);\nif( dat == null )\nthrow new DMLException(\"Non-existent output variable: \"+varname);\n@@ -141,9 +135,8 @@ public class ResultVariables\n* @param varname\n* output variable name\n* @return double value\n- * @throws DMLException if DMLException occurs\n*/\n- public double getDouble(String varname) throws DMLException {\n+ public double getDouble(String varname) {\nreturn getScalarObject(varname).getDoubleValue();\n}\n@@ -153,9 +146,8 @@ public class ResultVariables\n* @param varname\n* output variable name\n* @return boolean value\n- * @throws DMLException if DMLException occurs\n*/\n- public boolean getBoolean(String varname) throws DMLException {\n+ public boolean getBoolean(String varname) {\nreturn getScalarObject(varname).getBooleanValue();\n}\n@@ -165,9 +157,8 @@ public class ResultVariables\n* @param varname\n* output variable name\n* @return long value\n- * @throws DMLException if DMLException occurs\n*/\n- public long getLong(String varname) throws DMLException {\n+ public long getLong(String varname) {\nreturn getScalarObject(varname).getLongValue();\n}\n@@ -177,9 +168,8 @@ public class ResultVariables\n* @param varname\n* output variable name\n* @return string value\n- * @throws DMLException if DMLException occurs\n*/\n- public String getString(String varname) throws DMLException {\n+ public String getString(String varname) {\nreturn getScalarObject(varname).getStringValue();\n}\n@@ -189,9 +179,8 @@ public class ResultVariables\n* @param varname\n* output variable name\n* @return ScalarObject\n- * @throws DMLException if DMLException occurs\n*/\n- public ScalarObject getScalarObject(String varname) throws DMLException {\n+ public ScalarObject getScalarObject(String varname) {\nData dat = _out.get(varname);\nif( dat == null )\nthrow new DMLException(\"Non-existent output variable: \" + varname);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "diff": "@@ -32,7 +32,6 @@ import org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\nimport org.apache.log4j.Level;\nimport org.apache.log4j.Logger;\n-import org.apache.sysml.api.DMLException;\nimport org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\nimport org.apache.sysml.conf.ConfigurationManager;\n@@ -489,9 +488,7 @@ public class SpoofCompiler\n////////////////////\n// Codegen plan construction\n- private static void rExploreCPlans(Hop hop, CPlanMemoTable memo, boolean compileLiterals)\n- throws DMLException\n- {\n+ private static void rExploreCPlans(Hop hop, CPlanMemoTable memo, boolean compileLiterals) {\n//top-down memoization of processed dag nodes\nif( memo.contains(hop.getHopID()) || memo.containsHop(hop) )\nreturn;\n@@ -545,9 +542,7 @@ public class SpoofCompiler\nreturn P;\n}\n- private static void rConstructCPlans(Hop hop, CPlanMemoTable memo, HashMap<Long, Pair<Hop[],CNodeTpl>> cplans, boolean compileLiterals, HashSet<Long> visited)\n- throws DMLException\n- {\n+ private static void rConstructCPlans(Hop hop, CPlanMemoTable memo, HashMap<Long, Pair<Hop[],CNodeTpl>> cplans, boolean compileLiterals, HashSet<Long> visited) {\n//top-down memoization of processed dag nodes\nif( hop == null || visited.contains(hop.getHopID()) )\nreturn;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/ParUnaryAggregateTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/ParUnaryAggregateTest.java", "diff": "@@ -1092,7 +1092,7 @@ public class ParUnaryAggregateTest extends AutomatedTestBase\n|| aggtype == AggType.ROWMINS || aggtype == AggType.ROWMINS)?rows:1;\nint dim2 = (aggtype == AggType.COLSUMS || aggtype == AggType.COLSUMSSQ\n|| aggtype == AggType.COLMAXS || aggtype == AggType.COLMINS)?cols1:1;\n- TestUtils.compareMatrices(d1, d2, dim1, dim2, 0.00000000001);\n+ TestUtils.compareMatrices(d1, d2, dim1, dim2, 0.000000001);\n}\ncatch(Exception ex) {\nthrow new RuntimeException(ex);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -2443,7 +2443,7 @@ public class DMLTranslator\ncase EXISTS:\ncurrBuiltinOp = new UnaryOp(target.getName(), target.getDataType(),\n- target.getValueType(), Hop.OpOp1.EXISTS, new LiteralOp(expr.getName()));\n+ target.getValueType(), Hop.OpOp1.EXISTS, expr);\nbreak;\ncase SUM:\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/AggregateUnaryCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/AggregateUnaryCPInstruction.java", "diff": "@@ -122,9 +122,10 @@ public class AggregateUnaryCPInstruction extends UnaryCPInstruction\nec.setScalarOutput(output_name, new IntObject(rval));\n}\nelse if( _type == AUType.EXISTS ) {\n- //probe existing of variable in symbol table w/o error\n- boolean rval = ec.getVariables().keySet()\n- .contains(ec.getScalarInput(input1).getStringValue());\n+ //probe existence of variable in symbol table w/o error\n+ String varName = input1.isMatrix() ? input1.getName() :\n+ ec.getScalarInput(input1).getStringValue();\n+ boolean rval = ec.getVariables().keySet().contains(varName);\n//create and set output scalar\nec.setScalarOutput(output_name, new BooleanObject(rval));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/CPOperand.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/CPOperand.java", "diff": "@@ -69,6 +69,10 @@ public class CPOperand\nreturn _dataType.isMatrix();\n}\n+ public boolean isScalar() {\n+ return _dataType.isScalar();\n+ }\n+\npublic boolean isLiteral() {\nreturn _isLiteral;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/VariableCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/VariableCPInstruction.java", "diff": "@@ -802,12 +802,10 @@ public class VariableCPInstruction extends CPInstruction {\n*/\npublic static void processRemoveVariableInstruction( ExecutionContext ec, String varname ) {\n// remove variable from symbol table\n- Data input1_data = ec.removeVariable(varname);\n- if ( input1_data == null )\n- throw new DMLRuntimeException(\"Unexpected error: could not find a data object for variable name:\" + varname + \", while processing rmvar instruction.\");\n+ Data dat = ec.removeVariable(varname);\n//cleanup matrix data on fs/hdfs (if necessary)\n- if ( input1_data instanceof CacheableData ) {\n- ec.cleanupCacheableData( (CacheableData<?>) input1_data );\n+ if ( dat != null && dat instanceof CacheableData ) {\n+ ec.cleanupCacheableData((CacheableData<?>) dat);\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/ExistsVariableTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/ExistsVariableTest.java", "diff": "@@ -29,27 +29,39 @@ import org.apache.sysml.test.utils.TestUtils;\npublic class ExistsVariableTest extends AutomatedTestBase\n{\n- private final static String TEST_NAME1 = \"Exists\";\n+ private final static String TEST_NAME1 = \"Exists1\"; //for var names\n+ private final static String TEST_NAME2 = \"Exists2\"; //for vars\n+\nprivate final static String TEST_DIR = \"functions/misc/\";\nprivate final static String TEST_CLASS_DIR = TEST_DIR + ExistsVariableTest.class.getSimpleName() + \"/\";\n- //TODO additional test with variable creation in same DAG, requires better data dependency handling\n@Override\npublic void setUp() {\nTestUtils.clearAssertionInformation();\naddTestConfiguration( TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[]{\"R\"}));\n+ addTestConfiguration( TEST_NAME2, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME2, new String[]{\"R\"}));\n}\n@Test\n- public void testExistsPositive() {\n+ public void testExistsVarnamePositive() {\nrunExistsTest(TEST_NAME1, true);\n}\n@Test\n- public void testExistsNegative() {\n+ public void testExistsVarnameNegative() {\nrunExistsTest(TEST_NAME1, false);\n}\n+ @Test\n+ public void testExistsVarPositive() {\n+ runExistsTest(TEST_NAME2, true);\n+ }\n+\n+ @Test\n+ public void testExistsVarNegative() {\n+ runExistsTest(TEST_NAME2, false);\n+ }\n+\nprivate void runExistsTest(String testName, boolean pos) {\nTestConfiguration config = getTestConfiguration(testName);\nloadTestConfiguration(config);\n" }, { "change_type": "RENAME", "old_path": "src/test/scripts/functions/misc/Exists.dml", "new_path": "src/test/scripts/functions/misc/Exists1.dml", "diff": "" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/Exists2.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if( $1 == 1 )\n+ X = matrix(1,10,10);\n+\n+R = as.matrix(as.double(exists(X)));\n+\n+write(R, $2);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/unary/matrix/ExtractTriangularTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/unary/matrix/ExtractTriangularTest.java", "diff": "@@ -33,8 +33,8 @@ import org.apache.sysml.test.utils.TestUtils;\npublic class ExtractTriangularTest extends AutomatedTestBase\n{\n- private final static String TEST_NAME1 = \"extractLowerTri\";\n- private final static String TEST_NAME2 = \"extractUpperTri\";\n+ private final static String TEST_NAME1 = \"ExtractLowerTri\";\n+ private final static String TEST_NAME2 = \"ExtractUpperTri\";\nprivate final static String TEST_DIR = \"functions/unary/matrix/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + ExtractTriangularTest.class.getSimpleName() + \"/\";\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "diff": "@@ -291,7 +291,13 @@ public class SparkExecutionContext extends ExecutionContext\n@SuppressWarnings(\"unchecked\")\npublic JavaPairRDD<MatrixIndexes,MatrixBlock> getBinaryBlockRDDHandleForVariable( String varname ) {\nreturn (JavaPairRDD<MatrixIndexes,MatrixBlock>)\n- getRDDHandleForVariable( varname, InputInfo.BinaryBlockInputInfo);\n+ getRDDHandleForVariable( varname, InputInfo.BinaryBlockInputInfo, -1);\n+ }\n+\n+ @SuppressWarnings(\"unchecked\")\n+ public JavaPairRDD<MatrixIndexes,MatrixBlock> getBinaryBlockRDDHandleForVariable( String varname, int numParts ) {\n+ return (JavaPairRDD<MatrixIndexes,MatrixBlock>)\n+ getRDDHandleForVariable( varname, InputInfo.BinaryBlockInputInfo, numParts);\n}\n/**\n@@ -304,15 +310,19 @@ public class SparkExecutionContext extends ExecutionContext\n@SuppressWarnings(\"unchecked\")\npublic JavaPairRDD<Long,FrameBlock> getFrameBinaryBlockRDDHandleForVariable( String varname ) {\nJavaPairRDD<Long,FrameBlock> out = (JavaPairRDD<Long,FrameBlock>)\n- getRDDHandleForVariable( varname, InputInfo.BinaryBlockInputInfo);\n+ getRDDHandleForVariable( varname, InputInfo.BinaryBlockInputInfo, -1);\nreturn out;\n}\npublic JavaPairRDD<?,?> getRDDHandleForVariable( String varname, InputInfo inputInfo ) {\n+ return getRDDHandleForVariable(varname, inputInfo, -1);\n+ }\n+\n+ public JavaPairRDD<?,?> getRDDHandleForVariable( String varname, InputInfo inputInfo, int numParts ) {\nData dat = getVariable(varname);\nif( dat instanceof MatrixObject ) {\nMatrixObject mo = getMatrixObject(varname);\n- return getRDDHandleForMatrixObject(mo, inputInfo);\n+ return getRDDHandleForMatrixObject(mo, inputInfo, numParts);\n}\nelse if( dat instanceof FrameObject ) {\nFrameObject fo = getFrameObject(varname);\n@@ -323,16 +333,12 @@ public class SparkExecutionContext extends ExecutionContext\n}\n}\n- /**\n- * This call returns an RDD handle for a given matrix object. This includes\n- * the creation of RDDs for in-memory or binary-block HDFS data.\n- *\n- * @param mo matrix object\n- * @param inputInfo input info\n- * @return JavaPairRDD handle for a matrix object\n- */\n- @SuppressWarnings(\"unchecked\")\npublic JavaPairRDD<?,?> getRDDHandleForMatrixObject( MatrixObject mo, InputInfo inputInfo ) {\n+ return getRDDHandleForMatrixObject(mo, inputInfo, -1);\n+ }\n+\n+ @SuppressWarnings(\"unchecked\")\n+ public JavaPairRDD<?,?> getRDDHandleForMatrixObject( MatrixObject mo, InputInfo inputInfo, int numParts ) {\n//NOTE: MB this logic should be integrated into MatrixObject\n//However, for now we cannot assume that spark libraries are\n//always available and hence only store generic references in\n@@ -366,7 +372,7 @@ public class SparkExecutionContext extends ExecutionContext\n}\nelse { //default case\nMatrixBlock mb = mo.acquireRead(); //pin matrix in memory\n- rdd = toMatrixJavaPairRDD(sc, mb, (int)mo.getNumRowsPerBlock(), (int)mo.getNumColumnsPerBlock());\n+ rdd = toMatrixJavaPairRDD(sc, mb, (int)mo.getNumRowsPerBlock(), (int)mo.getNumColumnsPerBlock(), numParts);\nmo.release(); //unpin matrix\n_parRDDs.registerRDD(rdd.id(), OptimizerUtils.estimatePartitionedSizeExactSparsity(mc), true);\n}\n@@ -657,16 +663,11 @@ public class SparkExecutionContext extends ExecutionContext\nobj.setRDDHandle( rddhandle );\n}\n- /**\n- * Utility method for creating an RDD out of an in-memory matrix block.\n- *\n- * @param sc java spark context\n- * @param src matrix block\n- * @param brlen block row length\n- * @param bclen block column length\n- * @return JavaPairRDD handle to matrix block\n- */\npublic static JavaPairRDD<MatrixIndexes,MatrixBlock> toMatrixJavaPairRDD(JavaSparkContext sc, MatrixBlock src, int brlen, int bclen) {\n+ return toMatrixJavaPairRDD(sc, src, brlen, bclen, -1);\n+ }\n+\n+ public static JavaPairRDD<MatrixIndexes,MatrixBlock> toMatrixJavaPairRDD(JavaSparkContext sc, MatrixBlock src, int brlen, int bclen, int numParts) {\nlong t0 = DMLScript.STATISTICS ? System.nanoTime() : 0;\nList<Tuple2<MatrixIndexes,MatrixBlock>> list = null;\n@@ -681,7 +682,9 @@ public class SparkExecutionContext extends ExecutionContext\n.collect(Collectors.toList());\n}\n- JavaPairRDD<MatrixIndexes,MatrixBlock> result = sc.parallelizePairs(list);\n+ JavaPairRDD<MatrixIndexes,MatrixBlock> result = (numParts > 1) ?\n+ sc.parallelizePairs(list, numParts) : sc.parallelizePairs(list);\n+\nif (DMLScript.STATISTICS) {\nStatistics.accSparkParallelizeTime(System.nanoTime() - t0);\nStatistics.incSparkParallelizeCount(1);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "diff": "@@ -97,8 +97,11 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixCharacteristics mcRdd = sec.getMatrixCharacteristics(rddVar);\nMatrixCharacteristics mcBc = sec.getMatrixCharacteristics(bcastVar);\n- //get input rdd\n- JavaPairRDD<MatrixIndexes,MatrixBlock> in1 = sec.getBinaryBlockRDDHandleForVariable(rddVar);\n+ //get input rdd with preferred number of partitions to avoid unnecessary repartition\n+ JavaPairRDD<MatrixIndexes,MatrixBlock> in1 = sec.getBinaryBlockRDDHandleForVariable(rddVar,\n+ (requiresFlatMapFunction(type, mcBc) && requiresRepartitioning(\n+ type, mcRdd, mcBc, sec.getSparkContext().defaultParallelism())) ?\n+ getNumRepartitioning(type, mcRdd, mcBc) : -1);\n//investigate if a repartitioning - including a potential flip of broadcast and rdd\n//inputs - is required to ensure moderately sized output partitions (2GB limitation)\n@@ -216,7 +219,8 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nboolean isLargeOutput = (OptimizerUtils.estimatePartitionedSizeExactSparsity(isLeft?mcBc.getRows():mcRdd.getRows(),\nisLeft?mcRdd.getCols():mcBc.getCols(), isLeft?mcBc.getRowsPerBlock():mcRdd.getRowsPerBlock(),\nisLeft?mcRdd.getColsPerBlock():mcBc.getColsPerBlock(), 1.0) / numPartitions) > 1024*1024*1024;\n- return isOuter && isLargeOutput && mcRdd.dimsKnown() && mcBc.dimsKnown();\n+ return isOuter && isLargeOutput && mcRdd.dimsKnown() && mcBc.dimsKnown()\n+ && numPartitions < getNumRepartitioning(type, mcRdd, mcBc);\n}\n/**\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/CreateSparseBlockFunction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/CreateSparseBlockFunction.java", "diff": "@@ -45,7 +45,8 @@ public class CreateSparseBlockFunction implements Function<MatrixBlock,MatrixBlo\n{\n//convert given block to CSR representation if in sparse format\n//but allow shallow pass-through if already in CSR representation.\n- if( arg0.isInSparseFormat() && !(arg0 instanceof CompressedMatrixBlock) )\n+ if( arg0.isInSparseFormat() && !arg0.isEmptyBlock(false)\n+ && !(arg0 instanceof CompressedMatrixBlock) )\nreturn new MatrixBlock(arg0, _stype, false);\nelse //pass through dense\nreturn arg0;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CheckpointSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CheckpointSPInstruction.java", "diff": "@@ -41,6 +41,7 @@ import org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysml.runtime.matrix.data.SparseBlock;\nimport org.apache.sysml.runtime.matrix.operators.Operator;\n+import org.apache.sysml.runtime.util.UtilFunctions;\npublic class CheckpointSPInstruction extends UnarySPInstruction {\n// default storage level\n@@ -100,19 +101,28 @@ public class CheckpointSPInstruction extends UnarySPInstruction {\nJavaPairRDD<?,?> out = null;\nif( !in.getStorageLevel().equals( _level ) )\n{\n- //(trigger coalesce if intended number of partitions exceeded by 20%\n- //and not hash partitioned to avoid losing the existing partitioner)\n+ //determine need for coalesce or repartition, and csr conversion\nint numPartitions = SparkUtils.getNumPreferredPartitions(mcIn, in);\nboolean coalesce = ( 1.2*numPartitions < in.getNumPartitions()\n&& !SparkUtils.isHashPartitioned(in) && in.getNumPartitions()\n> SparkExecutionContext.getDefaultParallelism(true));\n+ boolean repartition = mcIn.dimsKnown(true) && mcIn.isUltraSparse()\n+ && numPartitions > in.getNumPartitions();\n+ boolean mcsr2csr = input1.getDataType()==DataType.MATRIX\n+ && OptimizerUtils.checkSparseBlockCSRConversion(mcIn)\n+ && !_level.equals(Checkpoint.SER_STORAGE_LEVEL);\n//checkpoint pre-processing rdd operations\nif( coalesce ) {\n//merge partitions without shuffle if too many partitions\nout = in.coalesce( numPartitions );\n}\n- else {\n+ else if( repartition ) {\n+ //repartition to preferred size as multiple of default parallelism\n+ out = in.repartition(UtilFunctions.roundToNext(numPartitions,\n+ SparkExecutionContext.getDefaultParallelism(true)));\n+ }\n+ else if( !mcsr2csr ) {\n//since persist is an in-place marker for a storage level, we\n//apply a narrow shallow copy to allow for short-circuit collects\nif( input1.getDataType() == DataType.MATRIX )\n@@ -122,11 +132,12 @@ public class CheckpointSPInstruction extends UnarySPInstruction {\nout = ((JavaPairRDD<Long,FrameBlock>)in)\n.mapValues(new CopyFrameBlockFunction(false));\n}\n+ else {\n+ out = in;\n+ }\n//convert mcsr into memory-efficient csr if potentially sparse\n- if( input1.getDataType()==DataType.MATRIX\n- && OptimizerUtils.checkSparseBlockCSRConversion(mcIn)\n- && !_level.equals(Checkpoint.SER_STORAGE_LEVEL) ) {\n+ if( mcsr2csr ) {\nout = ((JavaPairRDD<MatrixIndexes,MatrixBlock>)out)\n.mapValues(new CreateSparseBlockFunction(SparseBlock.Type.CSR));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/RDDAggregateUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/RDDAggregateUtils.java", "diff": "@@ -262,14 +262,16 @@ public class RDDAggregateUtils\npublic CorrMatrixBlock call(CorrMatrixBlock arg0, MatrixBlock arg1)\nthrows Exception\n{\n+ if( arg1.isEmptyBlock(false) )\n+ return arg0;\n+\n//get current block and correction\nMatrixBlock value = arg0.getValue();\nMatrixBlock corr = arg0.getCorrection();\n//correction block allocation on demand\n- if( corr == null ){\n+ if( corr == null )\ncorr = new MatrixBlock(value.getNumRows(), value.getNumColumns(), false);\n- }\n//aggregate other input and maintain corrections\n//(existing value and corr are used in place)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/MatrixCharacteristics.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/MatrixCharacteristics.java", "diff": "@@ -24,6 +24,7 @@ import java.io.Serializable;\nimport java.util.Arrays;\nimport java.util.HashMap;\n+import org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.lops.MMTSJ.MMTSJType;\nimport org.apache.sysml.runtime.instructions.mr.AggregateBinaryInstruction;\nimport org.apache.sysml.runtime.instructions.mr.AggregateInstruction;\n@@ -61,6 +62,7 @@ import org.apache.sysml.runtime.instructions.mr.UaggOuterChainInstruction;\nimport org.apache.sysml.runtime.instructions.mr.UnaryInstruction;\nimport org.apache.sysml.runtime.instructions.mr.UnaryMRInstructionBase;\nimport org.apache.sysml.runtime.instructions.mr.ZeroOutInstruction;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.ReorgOperator;\n@@ -225,6 +227,11 @@ public class MatrixCharacteristics implements Serializable\nreturn ( !ubNnz && nonZero >= 0 );\n}\n+ public boolean isUltraSparse() {\n+ return dimsKnown(true) && OptimizerUtils.getSparsity(this)\n+ < MatrixBlock.ULTRA_SPARSITY_TURN_POINT;\n+ }\n+\npublic boolean mightHaveEmptyBlocks() {\nlong singleBlk = Math.max(Math.min(numRows, numRowsPerBlock),1)\n* Math.max(Math.min(numColumns, numColumnsPerBlock),1);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -1512,17 +1512,18 @@ public class LibMatrixMult\nif( alen==1 ) {\n//row selection (now aggregation) with potential scaling\nint aix = aixs[apos];\n+ int lnnz = 0;\nif( rightSparse ) { //sparse right matrix (full row copy)\nif( !m2.sparseBlock.isEmpty(aix) ) {\nret.rlen=m;\nret.allocateSparseRowsBlock(false); //allocation on demand\nboolean ldeep = (m2.sparseBlock instanceof SparseBlockMCSR);\nret.sparseBlock.set(i, m2.sparseBlock.get(aix), ldeep);\n- ret.nonZeros += ret.sparseBlock.size(i);\n+ ret.nonZeros += (lnnz = ret.sparseBlock.size(i));\n}\n}\nelse { //dense right matrix (append all values)\n- int lnnz = (int)m2.recomputeNonZeros(aix, aix, 0, n-1);\n+ lnnz = (int)m2.recomputeNonZeros(aix, aix, 0, n-1);\nif( lnnz > 0 ) {\nc.allocate(i, lnnz); //allocate once\ndouble[] bvals = m2.getDenseBlock().values(aix);\n@@ -1532,7 +1533,7 @@ public class LibMatrixMult\n}\n}\n//optional scaling if not pure selection\n- if( avals[apos] != 1 )\n+ if( avals[apos] != 1 && lnnz > 0 )\nvectMultiplyInPlace(avals[apos], c.values(i), c.pos(i), c.size(i));\n}\nelse //GENERAL CASE\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/binary/matrix_full_other/FullMatrixMultiplicationUltraSparseTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/binary/matrix_full_other/FullMatrixMultiplicationUltraSparseTest.java", "diff": "@@ -24,7 +24,7 @@ import java.util.HashMap;\nimport org.junit.AfterClass;\nimport org.junit.BeforeClass;\nimport org.junit.Test;\n-\n+import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\nimport org.apache.sysml.lops.LopProperties.ExecType;\nimport org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex;\n@@ -34,7 +34,6 @@ import org.apache.sysml.test.utils.TestUtils;\npublic class FullMatrixMultiplicationUltraSparseTest extends AutomatedTestBase\n{\n-\nprivate final static String TEST_NAME = \"FullMatrixMultiplication\";\nprivate final static String TEST_DIR = \"functions/binary/matrix_full_other/\";\nprivate final static String TEST_CLASS_DIR = TEST_DIR + FullMatrixMultiplicationUltraSparseTest.class.getSimpleName() + \"/\";\n@@ -83,62 +82,77 @@ public class FullMatrixMultiplicationUltraSparseTest extends AutomatedTestBase\n}\n@Test\n- public void testMMDenseUltraSparseCP()\n- {\n+ public void testMMDenseUltraSparseCP() {\nrunMatrixMatrixMultiplicationTest(SparsityType.DENSE, SparsityType.ULTRA_SPARSE, ExecType.CP);\n}\n@Test\n- public void testMMSparseUltraSparseCP()\n- {\n+ public void testMMSparseUltraSparseCP() {\nrunMatrixMatrixMultiplicationTest(SparsityType.SPARSE, SparsityType.ULTRA_SPARSE, ExecType.CP);\n}\n@Test\n- public void testMMUltraSparseDenseCP()\n- {\n+ public void testMMUltraSparseDenseCP() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.DENSE, ExecType.CP);\n}\n@Test\n- public void testMMUltraSparseSparseCP()\n- {\n+ public void testMMUltraSparseSparseCP() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.SPARSE, ExecType.CP);\n}\n@Test\n- public void testMMUltraSparseUltraSparseCP()\n- {\n+ public void testMMUltraSparseUltraSparseCP() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.ULTRA_SPARSE, ExecType.CP);\n}\n@Test\n- public void testMMDenseUltraSparseMR()\n- {\n+ public void testMMDenseUltraSparseSP() {\n+ runMatrixMatrixMultiplicationTest(SparsityType.DENSE, SparsityType.ULTRA_SPARSE, ExecType.SPARK);\n+ }\n+\n+ @Test\n+ public void testMMSparseUltraSparseSP() {\n+ runMatrixMatrixMultiplicationTest(SparsityType.SPARSE, SparsityType.ULTRA_SPARSE, ExecType.SPARK);\n+ }\n+\n+ @Test\n+ public void testMMUltraSparseDenseSP() {\n+ runMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.DENSE, ExecType.SPARK);\n+ }\n+\n+ @Test\n+ public void testMMUltraSparseSparseSP() {\n+ runMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.SPARSE, ExecType.SPARK);\n+ }\n+\n+ @Test\n+ public void testMMUltraSparseUltraSparseSP() {\n+ runMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.ULTRA_SPARSE, ExecType.SPARK);\n+ }\n+\n+ @Test\n+ public void testMMDenseUltraSparseMR() {\nrunMatrixMatrixMultiplicationTest(SparsityType.DENSE, SparsityType.ULTRA_SPARSE, ExecType.MR);\n}\n@Test\n- public void testMMSparseUltraSparseMR()\n- {\n+ public void testMMSparseUltraSparseMR() {\nrunMatrixMatrixMultiplicationTest(SparsityType.SPARSE, SparsityType.ULTRA_SPARSE, ExecType.MR);\n}\n@Test\n- public void testMMUltraSparseDenseMR()\n- {\n+ public void testMMUltraSparseDenseMR() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.DENSE, ExecType.MR);\n}\n@Test\n- public void testMMUltraSparseSparseMR()\n- {\n+ public void testMMUltraSparseSparseMR() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.SPARSE, ExecType.MR);\n}\n@Test\n- public void testMMUltraSparseUltraSparseMR()\n- {\n+ public void testMMUltraSparseUltraSparseMR() {\nrunMatrixMatrixMultiplicationTest(SparsityType.ULTRA_SPARSE, SparsityType.ULTRA_SPARSE, ExecType.MR);\n}\n@@ -150,11 +164,16 @@ public class FullMatrixMultiplicationUltraSparseTest extends AutomatedTestBase\n*/\nprivate void runMatrixMatrixMultiplicationTest( SparsityType sparseM1, SparsityType sparseM2, ExecType instType)\n{\n- //setup exec type, rows, cols\n-\n- //rtplatform for MR\nRUNTIME_PLATFORM platformOld = rtplatform;\n- rtplatform = (instType==ExecType.MR) ? RUNTIME_PLATFORM.HADOOP : RUNTIME_PLATFORM.HYBRID;\n+ switch( instType ){\n+ case MR: rtplatform = RUNTIME_PLATFORM.HADOOP; break;\n+ case SPARK: rtplatform = RUNTIME_PLATFORM.SPARK; break;\n+ default: rtplatform = RUNTIME_PLATFORM.HYBRID; break;\n+ }\n+\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK )\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\ntry\n{\n@@ -194,8 +213,8 @@ public class FullMatrixMultiplicationUltraSparseTest extends AutomatedTestBase\nHashMap<CellIndex, Double> rfile = readRMatrixFromFS(\"C\");\nTestUtils.compareMatrices(dmlfile, rfile, eps, \"Stat-DML\", \"Stat-R\");\n}\n- finally\n- {\n+ finally {\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = sparkConfigOld;\nrtplatform = platformOld;\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/ExtractBlockForBinaryReblock.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/ExtractBlockForBinaryReblock.java", "diff": "@@ -70,28 +70,26 @@ public class ExtractBlockForBinaryReblock implements PairFlatMapFunction<Tuple2<\nlong endRowGlobalCellIndex = getEndGlobalIndex(ixIn.getRowIndex(), true, true);\nlong startColGlobalCellIndex = UtilFunctions.computeCellIndex(ixIn.getColumnIndex(), in_bclen, 0);\nlong endColGlobalCellIndex = getEndGlobalIndex(ixIn.getColumnIndex(), true, false);\n- assert(startRowGlobalCellIndex <= endRowGlobalCellIndex && startColGlobalCellIndex <= endColGlobalCellIndex);\nlong out_startRowBlockIndex = UtilFunctions.computeBlockIndex(startRowGlobalCellIndex, out_brlen);\nlong out_endRowBlockIndex = UtilFunctions.computeBlockIndex(endRowGlobalCellIndex, out_brlen);\nlong out_startColBlockIndex = UtilFunctions.computeBlockIndex(startColGlobalCellIndex, out_bclen);\nlong out_endColBlockIndex = UtilFunctions.computeBlockIndex(endColGlobalCellIndex, out_bclen);\n- assert(out_startRowBlockIndex <= out_endRowBlockIndex && out_startColBlockIndex <= out_endColBlockIndex);\nArrayList<Tuple2<MatrixIndexes, MatrixBlock>> retVal = new ArrayList<>();\nfor(long i = out_startRowBlockIndex; i <= out_endRowBlockIndex; i++) {\nfor(long j = out_startColBlockIndex; j <= out_endColBlockIndex; j++) {\nMatrixIndexes indx = new MatrixIndexes(i, j);\n- long rowLower = Math.max(UtilFunctions.computeCellIndex(i, out_brlen, 0), startRowGlobalCellIndex);\n- long rowUpper = Math.min(getEndGlobalIndex(i, false, true), endRowGlobalCellIndex);\n- long colLower = Math.max(UtilFunctions.computeCellIndex(j, out_bclen, 0), startColGlobalCellIndex);\n- long colUpper = Math.min(getEndGlobalIndex(j, false, false), endColGlobalCellIndex);\n-\nint new_lrlen = UtilFunctions.computeBlockSize(rlen, i, out_brlen);\nint new_lclen = UtilFunctions.computeBlockSize(clen, j, out_bclen);\nMatrixBlock blk = new MatrixBlock(new_lrlen, new_lclen, true);\n+ if( !in.isEmptyBlock(false) ) {\n+ long rowLower = Math.max(UtilFunctions.computeCellIndex(i, out_brlen, 0), startRowGlobalCellIndex);\n+ long rowUpper = Math.min(getEndGlobalIndex(i, false, true), endRowGlobalCellIndex);\n+ long colLower = Math.max(UtilFunctions.computeCellIndex(j, out_bclen, 0), startColGlobalCellIndex);\n+ long colUpper = Math.min(getEndGlobalIndex(j, false, false), endColGlobalCellIndex);\nint in_i1 = UtilFunctions.computeCellInBlock(rowLower, in_brlen);\nint out_i1 = UtilFunctions.computeCellInBlock(rowLower, out_brlen);\n@@ -99,10 +97,11 @@ public class ExtractBlockForBinaryReblock implements PairFlatMapFunction<Tuple2<\nint in_j1 = UtilFunctions.computeCellInBlock(colLower, in_bclen);\nint out_j1 = UtilFunctions.computeCellInBlock(colLower, out_bclen);\nfor(long j1 = colLower; j1 <= colUpper; j1++, in_j1++, out_j1++) {\n- double val = in.getValue(in_i1, in_j1);\n+ double val = in.quickGetValue(in_i1, in_j1);\nblk.appendValue(out_i1, out_j1, val);\n}\n}\n+ }\nretVal.add(new Tuple2<>(indx, blk));\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -184,11 +184,13 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\nthrow new RuntimeException(\"Sparse matrix block expected.\");\n//deep copy and change sparse block type\n+ if( !that.isEmptyBlock(false) ) {\nnonZeros = that.nonZeros;\nestimatedNNzsPerRow = that.estimatedNNzsPerRow;\nsparseBlock = SparseBlockFactory\n.copySparseBlock(stype, that.sparseBlock, deep);\n}\n+ }\n////////\n// Initialization methods\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseBlockCSR.java", "diff": "@@ -734,6 +734,8 @@ public class SparseBlockCSR extends SparseBlock\n@Override\npublic double get(int r, int c) {\n+ if( isEmpty(r) )\n+ return 0;\nint pos = pos(r);\nint len = size(r);\n@@ -744,6 +746,8 @@ public class SparseBlockCSR extends SparseBlock\n@Override\npublic SparseRow get(int r) {\n+ if( isEmpty(r) )\n+ return new SparseRowScalar();\nint pos = pos(r);\nint len = size(r);\n@@ -751,7 +755,6 @@ public class SparseBlockCSR extends SparseBlock\nSystem.arraycopy(_indexes, pos, row.indexes(), 0, len);\nSystem.arraycopy(_values, pos, row.values(), 0, len);\nrow.setSize(len);\n-\nreturn row;\n}\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteNNIssueTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.test.integration.functions.misc;\n+\n+import org.junit.Test;\n+\n+import java.util.HashMap;\n+\n+import org.apache.sysml.hops.OptimizerUtils;\n+import org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.integration.TestConfiguration;\n+import org.apache.sysml.test.utils.TestUtils;\n+\n+public class RewriteNNIssueTest extends AutomatedTestBase\n+{\n+ private static final String TEST_NAME = \"RewriteNNIssue\";\n+\n+ private static final String TEST_DIR = \"functions/misc/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + RewriteNNIssueTest.class.getSimpleName() + \"/\";\n+\n+ private double eps = Math.pow(10, -10);\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration( TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[] { \"R\" }) );\n+ }\n+\n+ @Test\n+ public void testNNIssueRewrite() {\n+ runNNIssueTest(true);\n+ }\n+\n+ @Test\n+ public void testNNIssueNoRewrite() {\n+ runNNIssueTest(false);\n+ }\n+\n+ private void runNNIssueTest(boolean rewrites)\n+ {\n+ boolean oldFlag = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n+\n+ try {\n+ TestConfiguration config = getTestConfiguration(TEST_NAME);\n+ loadTestConfiguration(config);\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[]{ \"-stats\",\"-args\", output(\"R\") };\n+ fullRScriptName = HOME + TEST_NAME + \".R\";\n+ rCmd = getRCmd(expectedDir());\n+\n+ OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION = rewrites;\n+\n+ //run test\n+ runTest(true, false, null, -1);\n+ runRScript(true);\n+\n+ //compare matrices\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(\"R\");\n+ HashMap<CellIndex, Double> rfile = readRMatrixFromFS(\"R\");\n+ TestUtils.compareMatrices(dmlfile, rfile, eps, \"Stat-DML\", \"Stat-R\");\n+ }\n+ finally {\n+ OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION = oldFlag;\n+ }\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/RewriteNNIssue.R", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(Matrix)\n+library(matrixStats)\n+\n+N = 2\n+C = 2\n+Hin = 3\n+Win = 4\n+\n+X = matrix(cbind(seq(1,20),seq(1,20),seq(1,8)), nrow=2, ncol=24, byrow=TRUE)\n+gamma = matrix(c(1,2), byrow=TRUE, nrow=2, ncol=1)\n+beta = matrix(c(0,1), byrow=TRUE, nrow=2, ncol=1)\n+ema_mean = matrix(c(4,5), byrow=TRUE, nrow=2, ncol=1)\n+ema_var = matrix(c(2,3), byrow=TRUE, nrow=2, ncol=1)\n+mu = 0.95\n+epsilon = 1e-4\n+\n+subgrp_means = matrix(colMeans(X), nrow=C, ncol=Hin*Win, byrow=TRUE)\n+subgrp_vars = matrix(colVars(X) * ((N-1)/N), nrow=C, ncol=Hin*Win, byrow=TRUE)\n+mean = rowMeans(subgrp_means) # shape (C, 1)\n+var = rowMeans(subgrp_vars) + rowVars(subgrp_means)*(((Hin*Win)-1)/(Hin*Win))\n+ema_mean_upd = mu*ema_mean + (1-mu)*mean\n+ema_var_upd = mu*ema_var + (1-mu)*var\n+\n+R = cbind(mean, var, ema_mean_upd, ema_var_upd)\n+\n+writeMM(as(R, \"CsparseMatrix\"), paste(args[1], \"R\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/RewriteNNIssue.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+N = 2\n+C = 2\n+Hin = 3\n+Win = 4\n+X = matrix(rbind(seq(1,20),seq(1,20),seq(1,8)), rows=2, cols=24)\n+gamma = matrix(\"1 2\", rows=2, cols=1)\n+beta = matrix(\"0 1\", rows=2, cols=1)\n+ema_mean = matrix(\"4 5\", rows=2, cols=1)\n+ema_var = matrix(\"2 3\", rows=2, cols=1)\n+mu = 0.95\n+epsilon = 1e-4\n+\n+subgrp_means = matrix(colMeans(X), rows=C, cols=Hin*Win)\n+subgrp_vars = matrix(colVars(X) * ((N-1)/N), rows=C, cols=Hin*Win)\n+mean = rowMeans(subgrp_means) # shape (C, 1)\n+var = rowMeans(subgrp_vars) + rowVars(subgrp_means)*(((Hin*Win)-1)/(Hin*Win))\n+ema_mean_upd = mu*ema_mean + (1-mu)*mean\n+ema_var_upd = mu*ema_var + (1-mu)*var\n+\n+R = cbind(mean, var, ema_mean_upd, ema_var_upd)\n+\n+write(R,$1)\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/misc/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/misc/ZPackageSuite.java", "diff": "@@ -65,6 +65,7 @@ import org.junit.runners.Suite;\nRewriteLoopVectorization.class,\nRewriteMatrixMultChainOptTest.class,\nRewriteMergeBlocksTest.class,\n+ RewriteNNIssueTest.class,\nRewritePushdownSumBinaryMult.class,\nRewritePushdownSumOnBinaryTest.class,\nRewritePushdownUaggTest.class,\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/Connection.java", "diff": "@@ -421,7 +421,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic double[][] convertToDoubleMatrix(String input, int rows, int cols) throws IOException {\n- return convertToDoubleMatrix(IOUtilFunctions.toInputStream(input), rows, cols);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToDoubleMatrix(is, rows, cols);\n+ }\n}\n/**\n@@ -467,7 +469,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic MatrixBlock convertToMatrix(String input, String meta) throws IOException {\n- return convertToMatrix(IOUtilFunctions.toInputStream(input), meta);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToMatrix(is, meta);\n+ }\n}\n/**\n@@ -509,7 +513,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic MatrixBlock convertToMatrix(String input, int rows, int cols) throws IOException {\n- return convertToMatrix(IOUtilFunctions.toInputStream(input), rows, cols);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToMatrix(is, rows, cols);\n+ }\n}\n/**\n@@ -655,7 +661,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic String[][] convertToStringFrame(String input, int rows, int cols) throws IOException {\n- return convertToStringFrame(IOUtilFunctions.toInputStream(input), rows, cols);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToStringFrame(is, rows, cols);\n+ }\n}\n/**\n@@ -701,7 +709,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic FrameBlock convertToFrame(String input, String meta) throws IOException {\n- return convertToFrame(IOUtilFunctions.toInputStream(input), meta);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToFrame(is, meta);\n+ }\n}\n/**\n@@ -743,7 +753,9 @@ public class Connection implements Closeable\n* @throws IOException if IOException occurs\n*/\npublic FrameBlock convertToFrame(String input, int rows, int cols) throws IOException {\n- return convertToFrame(IOUtilFunctions.toInputStream(input), rows, cols);\n+ try( InputStream is = IOUtilFunctions.toInputStream(input) ) {\n+ return convertToFrame(is, rows, cols);\n+ }\n}\n/**\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptFactory.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptFactory.java", "diff": "@@ -298,8 +298,11 @@ public class ScriptFactory {\nif (!resourcePath.startsWith(\"/\")) {\nresourcePath = \"/\" + resourcePath;\n}\n- InputStream inputStream = ScriptFactory.class.getResourceAsStream(resourcePath);\n+ try( InputStream inputStream = ScriptFactory.class.getResourceAsStream(resourcePath) ) {\nreturn scriptFromInputStream(inputStream, scriptType).setName(resourcePath);\n+ } catch (Exception e){\n+ throw new MLContextException(\"Error trying to read script from resource: \"+ resourcePath, e);\n+ }\n}\n/**\n@@ -425,8 +428,7 @@ public class ScriptFactory {\nif ((!urlString.toLowerCase().startsWith(\"http:\")) && (!urlString.toLowerCase().startsWith(\"https:\"))) {\nthrow new MLContextException(\"Currently only reading from http and https URLs is supported\");\n}\n- try {\n- InputStream is = url.openStream();\n+ try( InputStream is = url.openStream() ) {\nreturn IOUtils.toString(is);\n} catch (IOException e) {\nthrow new MLContextException(\"Error trying to read script string from URL: \" + url, e);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/ParserWrapper.java", "new_path": "src/main/java/org/apache/sysml/parser/ParserWrapper.java", "diff": "@@ -124,7 +124,9 @@ public abstract class ParserWrapper {\n{\nString resPath = scriptPathToResourcePath(script);\nLOG.debug(\"Looking for the following resource from the SystemML jar file: \" + resPath);\n- InputStream is = ParserWrapper.class.getResourceAsStream(resPath);\n+ InputStream is = null;\n+ try {\n+ is = ParserWrapper.class.getResourceAsStream(resPath);\nif (is == null) {\nif (resPath.startsWith(\"/scripts\")) {\nLOG.error(\"Failed to read from the file system ('\" + script + \"') or SystemML jar file ('\" + resPath + \"')\");\n@@ -143,6 +145,10 @@ public abstract class ParserWrapper {\nString s = IOUtils.toString(is);\nreturn s;\n}\n+ finally {\n+ IOUtilFunctions.closeSilently(is);\n+ }\n+ }\nfinally {\nIOUtilFunctions.closeSilently(in);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/JCudaKernels.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/JCudaKernels.java", "diff": "@@ -125,10 +125,8 @@ public class JCudaKernels {\n* @return\n*/\nprivate static Pointer initKernels(String ptxFileName) {\n- InputStream in = null;\nByteArrayOutputStream out = null;\n- try {\n- in = JCudaKernels.class.getResourceAsStream(ptxFileName);\n+ try( InputStream in = JCudaKernels.class.getResourceAsStream(ptxFileName) ) {\nif (in != null) {\nout = new ByteArrayOutputStream();\nbyte buffer[] = new byte[8192];\n@@ -149,7 +147,6 @@ public class JCudaKernels {\nthrow new DMLRuntimeException(\"Could not initialize the kernels\", e);\n} finally {\nIOUtilFunctions.closeSilently(out);\n- IOUtilFunctions.closeSilently(in);\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/transform/meta/TfMetaUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/transform/meta/TfMetaUtils.java", "diff": "@@ -216,7 +216,7 @@ public class TfMetaUtils\nthrows IOException\n{\n//read column names\n- String colnamesStr = IOUtilFunctions.toString(Connection.class.getResourceAsStream(metapath+\"/\"+TfUtils.TXMTD_COLNAMES));\n+ String colnamesStr = getStringFromResource(metapath+\"/\"+TfUtils.TXMTD_COLNAMES);\nString[] colnames = IOUtilFunctions.split(colnamesStr.trim(), colDelim);\n//read meta data (currently supported: recode, dummycode, bin, omit)\n@@ -228,22 +228,22 @@ public class TfMetaUtils\nString colName = colnames[j];\n//read recode maps for recoded or dummycoded columns\nString name = metapath+\"/\"+\"Recode\"+\"/\"+colName;\n- String map = IOUtilFunctions.toString(Connection.class.getResourceAsStream(name+TfUtils.TXMTD_RCD_MAP_SUFFIX));\n+ String map = getStringFromResource(name+TfUtils.TXMTD_RCD_MAP_SUFFIX);\nif( map != null ) {\nmeta.put(colName, map);\n- String ndistinct = IOUtilFunctions.toString(Connection.class.getResourceAsStream(name+TfUtils.TXMTD_RCD_DISTINCT_SUFFIX));\n+ String ndistinct = getStringFromResource(name+TfUtils.TXMTD_RCD_DISTINCT_SUFFIX);\nrows = Math.max(rows, Integer.parseInt(ndistinct));\n}\n//read binning map for binned columns\nString name2 = metapath+\"/\"+\"Bin\"+\"/\"+colName;\n- String map2 = IOUtilFunctions.toString(Connection.class.getResourceAsStream(name2+TfUtils.TXMTD_BIN_FILE_SUFFIX));\n+ String map2 = getStringFromResource(name2+TfUtils.TXMTD_BIN_FILE_SUFFIX);\nif( map2 != null ) {\nmeta.put(colName, map2);\nrows = Math.max(rows, Integer.parseInt(map2.split(TfUtils.TXMTD_SEP)[4]));\n}\n//read impute value for mv columns\nString name3 = metapath+File.separator+\"Impute\"+File.separator+colName;\n- String map3 = IOUtilFunctions.toString(Connection.class.getResourceAsStream(name3+TfUtils.TXMTD_MV_FILE_SUFFIX));\n+ String map3 = getStringFromResource(name3+TfUtils.TXMTD_MV_FILE_SUFFIX);\nif( map3 != null ) {\nmvmeta.put(colName, map3);\n}\n@@ -389,4 +389,10 @@ public class TfMetaUtils\nthrow new IOException(ex);\n}\n}\n+\n+ private static String getStringFromResource(String path) throws IOException {\n+ try(InputStream is = Connection.class.getResourceAsStream(path) ) {\n+ return IOUtilFunctions.toString(is);\n+ }\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/NativeHelper.java", "new_path": "src/main/java/org/apache/sysml/utils/NativeHelper.java", "diff": "@@ -37,6 +37,7 @@ import org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.conf.DMLConfig;\nimport org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n+import org.apache.sysml.runtime.io.IOUtilFunctions;\n/**\n* This class helps in loading native library.\n@@ -291,33 +292,25 @@ public class NativeHelper {\nprivate static boolean loadLibraryHelper(String path) {\n- InputStream in = null; OutputStream out = null;\n- try {\n+ OutputStream out = null;\n+ try( InputStream in = NativeHelper.class.getResourceAsStream(\"/lib/\"+path) ) {\n// This logic is added because Java does not allow to load library from a resource file.\n- in = NativeHelper.class.getResourceAsStream(\"/lib/\"+path);\nif(in != null) {\nFile temp = File.createTempFile(path, \"\");\ntemp.deleteOnExit();\nout = FileUtils.openOutputStream(temp);\nIOUtils.copy(in, out);\n- in.close(); in = null;\n- out.close(); out = null;\nSystem.load(temp.getAbsolutePath());\nreturn true;\n}\nelse\nLOG.warn(\"No lib available in the jar:\" + path);\n- } catch(IOException e) {\n+ }\n+ catch(IOException e) {\nLOG.warn(\"Unable to load library \" + path + \" from resource:\" + e.getMessage());\n- } finally {\n- if(out != null)\n- try {\n- out.close();\n- } catch (IOException e) {}\n- if(in != null)\n- try {\n- in.close();\n- } catch (IOException e) {}\n+ }\n+ finally {\n+ IOUtilFunctions.closeSilently(out);\n}\nreturn false;\n}\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/parser/BuiltinConstant.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.parser;\n+\n+import org.apache.commons.lang3.EnumUtils;\n+\n+/**\n+ * These are the builtin constants\n+ */\n+public enum BuiltinConstant {\n+ PI(Math.PI),\n+ INF(Double.POSITIVE_INFINITY),\n+ NaN(Double.NaN);\n+\n+ private DoubleIdentifier _id;\n+\n+ private BuiltinConstant(double d) {\n+ this._id = new DoubleIdentifier(d);\n+ }\n+\n+ public DoubleIdentifier get() {\n+ return this._id;\n+ }\n+\n+ public static boolean contains(String name) {\n+ return EnumUtils.isValidEnum(BuiltinConstant.class, name);\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "new_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "diff": "@@ -784,6 +784,12 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nDataIdentifier target = as.getTarget();\nExpression source = as.getSource();\n+ // check if target is builtin constant\n+ if (target != null && BuiltinConstant.contains(target.getName())) {\n+ target.raiseValidateError(String.format(\n+ \"Cannot assign a value to the builtin constant %s.\", target.getName()), false);\n+ }\n+\nif (source instanceof FunctionCallIdentifier) {\n((FunctionCallIdentifier) source).validateExpression(\ndmlProg, ids.getVariables(),currConstVars, conditional);\n@@ -900,6 +906,13 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nArrayList<DataIdentifier> targetList = mas.getTargetList();\nExpression source = mas.getSource();\n+ // check if target list contains builtin constant\n+ targetList.forEach(target -> {\n+ if (target != null && BuiltinConstant.contains(target.getName()))\n+ target.raiseValidateError(String.format(\n+ \"Cannot assign a value to the builtin constant %s.\", target.getName()), false);\n+ });\n+\n//MultiAssignmentStatments currently supports only External,\n//User-defined, and Multi-return Builtin function expressions\nif (!(source instanceof DataIdentifier)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/common/CommonSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/common/CommonSyntacticValidator.java", "diff": "@@ -55,6 +55,7 @@ import org.apache.sysml.parser.Statement;\nimport org.apache.sysml.parser.StringIdentifier;\nimport org.apache.sysml.parser.dml.DmlSyntacticValidator;\nimport org.apache.sysml.parser.pydml.PydmlSyntacticValidator;\n+import org.apache.sysml.parser.BuiltinConstant;\n/**\n* Contains fields and (helper) methods common to {@link DmlSyntacticValidator} and {@link PydmlSyntacticValidator}\n@@ -328,6 +329,13 @@ public abstract class CommonSyntacticValidator {\n}\nprotected void exitDataIdExpressionHelper(ParserRuleContext ctx, ExpressionInfo me, ExpressionInfo dataInfo) {\n+ // inject builtin constant\n+ if (dataInfo.expr instanceof DataIdentifier) {\n+ DataIdentifier id = ((DataIdentifier) dataInfo.expr);\n+ if (BuiltinConstant.contains(id.getName())) {\n+ dataInfo.expr = BuiltinConstant.valueOf(id.getName()).get();\n+ }\n+ }\nme.expr = dataInfo.expr;\n// If \"The parameter $X either needs to be passed through commandline or initialized to default value\" validation\n// error occurs, then dataInfo.expr is null which would cause a null pointer exception with the following code.\n@@ -430,7 +438,7 @@ public abstract class CommonSyntacticValidator {\ninfo.stmt = new AssignmentStatement(ctx, target, source, currentFile);\n} catch (LanguageException e) {\n// TODO: extract more meaningful info from this exception.\n- notifyErrorListeners(\"invalid assignment\", lhsStart);\n+ notifyErrorListeners(\"invalid assignment: \" + e.getMessage(), lhsStart);\nreturn;\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "diff": "@@ -89,6 +89,14 @@ import scala.collection.Seq;\npublic class MLContextTest extends MLContextTestBase {\n+ @Test\n+ public void testBuiltinConstantsTest() {\n+ System.out.println(\"MLContextTest - basic builtin constants test\");\n+ Script script = dmlFromFile(baseDirectory + File.separator + \"builtin-constants-test.dml\");\n+ ml.execute(script);\n+ Assert.assertTrue(Statistics.getNoOfExecutedSPInst() == 0);\n+ }\n+\n@Test\npublic void testBasicExecuteEvalTest() {\nSystem.out.println(\"MLContextTest - basic eval test\");\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/org/apache/sysml/api/mlcontext/builtin-constants-test.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+# a func using builtin constant PI\n+f1 = function (double r) return (double res) {\n+ res = PI * r * r\n+}\n+\n+# use builtin constant PI in main func\n+res = PI * 1000\n+print(res)\n+\n+# use builtin constant NaN\n+print(NaN + 1)\n+\n+# use builtin constant INF\n+if (1 / 0 == INF) {\n+ print(\"1 / 0 is an infinity.\")\n+}\n+\n+print(f1(100))\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/scripts/applications/id3/id3.R", "new_path": "src/test/scripts/applications/id3/id3.R", "diff": "@@ -87,9 +87,9 @@ id3_learn = function(X, y, X_subset, attributes, minsplit)\nhist_labels1 = as.matrix(hist_labels1_helper[,2])\nnum_samples1 = sum(Tj)\nzero_entries_in_hist = (hist_labels1 == 0)\n- pi = hist_labels1/num_samples1\n- log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*pi\n- entropy_vector = -pi*log(log_term)\n+ piv = hist_labels1/num_samples1\n+ log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*piv\n+ entropy_vector = -piv*log(log_term)\nentropy = sum(entropy_vector)\nhxt_vector[j,1] = sum(Tj)/sum(X_subset)*entropy\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/applications/id3/id3.dml", "new_path": "src/test/scripts/applications/id3/id3.dml", "diff": "@@ -150,9 +150,9 @@ id3_learn = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] X_subset\nhist_labels1 = aggregate(target=Tj, groups=y, fn=\"sum\")\nnum_samples1 = sum(Tj)\nzero_entries_in_hist = (hist_labels1 == 0)\n- pi = hist_labels1/num_samples1\n- log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*pi\n- entropy_vector = -pi*log(log_term)\n+ piv = hist_labels1/num_samples1\n+ log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*piv\n+ entropy_vector = -piv*log(log_term)\nentropy = sum(entropy_vector)\nhxt_vector[j,1] = sum(Tj)/sum(X_subset)*entropy\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/applications/id3/id3.pydml", "new_path": "src/test/scripts/applications/id3/id3.pydml", "diff": "@@ -150,9 +150,9 @@ def id3_learn(X:matrix[float], y:matrix[float], X_subset:matrix[float], attribut\nhist_labels1 = aggregate(target=Tj, groups=y, fn=\"sum\")\nnum_samples1 = sum(Tj)\nzero_entries_in_hist = (hist_labels1 == 0)\n- pi = hist_labels1/num_samples1\n- log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*pi\n- entropy_vector = -pi*log(log_term)\n+ piv = hist_labels1/num_samples1\n+ log_term = zero_entries_in_hist*1 + (1-zero_entries_in_hist)*piv\n+ entropy_vector = -piv*log(log_term)\nentropy = sum(entropy_vector)\nhxt_vector[j-1,0] = sum(Tj)/sum(X_subset)*entropy\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -2611,6 +2611,7 @@ public class DMLTranslator\ncurrBuiltinOp = (expr2 == null) ? new UnaryOp(target.getName(), target.getDataType(), target.getValueType(),\nOpOp1.valueOf(source.getOpCode().name()), expr) : new BinaryOp(target.getName(), target.getDataType(),\ntarget.getValueType(), OpOp2.valueOf(source.getOpCode().name()), expr, expr2);\n+ break;\ncase IFELSE:\ncurrBuiltinOp=new TernaryOp(target.getName(), target.getDataType(), target.getValueType(),\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "diff": "@@ -1096,7 +1096,8 @@ public class BuiltinFunctionExpression extends DataIdentifier\nraiseValidateError(\"Second input to solve() must be a vector\", conditional);\nif ( getFirstExpr().getOutput().dimsKnown() && getSecondExpr().getOutput().dimsKnown() &&\n- getFirstExpr().getOutput().getDim1() != getSecondExpr().getOutput().getDim1() )\n+ getFirstExpr().getOutput().getDim1() != getSecondExpr().getOutput().getDim1() &&\n+ getFirstExpr().getOutput().getDim1() != getFirstExpr().getOutput().getDim2())\nraiseValidateError(\"Dimension mismatch in a call to solve()\", conditional);\noutput.setDataType(DataType.MATRIX);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "diff": "@@ -78,8 +78,11 @@ public class LibCommonsMath\n}\npublic static MatrixBlock matrixMatrixOperations(MatrixObject in1, MatrixObject in2, String opcode) {\n- if(opcode.equals(\"solve\"))\n+ if(opcode.equals(\"solve\")) {\n+ if (in1.getNumRows() != in1.getNumColumns())\n+ throw new DMLRuntimeException(\"The A matrix, in solve(A,b) should have squared dimensions.\");\nreturn computeSolve(in1, in2);\n+ }\nreturn null;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.spark;\nimport org.apache.spark.api.java.JavaPairRDD;\n+import org.apache.spark.api.java.JavaRDD;\n+import org.apache.spark.api.java.function.Function;\nimport org.apache.spark.api.java.function.PairFunction;\nimport scala.Tuple2;\n@@ -30,9 +32,12 @@ import org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.controlprogram.context.SparkExecutionContext;\nimport org.apache.sysml.runtime.functionobjects.Multiply;\nimport org.apache.sysml.runtime.functionobjects.Plus;\n+import org.apache.sysml.runtime.functionobjects.SwapIndex;\nimport org.apache.sysml.runtime.instructions.InstructionUtils;\nimport org.apache.sysml.runtime.instructions.cp.CPOperand;\nimport org.apache.sysml.runtime.instructions.spark.functions.FilterNonEmptyBlocksFunction;\n+import org.apache.sysml.runtime.instructions.spark.functions.FilterNonEmptyBlocksFunction2;\n+import org.apache.sysml.runtime.instructions.spark.functions.ReorgMapFunction;\nimport org.apache.sysml.runtime.instructions.spark.utils.RDDAggregateUtils;\nimport org.apache.sysml.runtime.instructions.spark.utils.SparkUtils;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n@@ -42,6 +47,7 @@ import org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;\nimport org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\nimport org.apache.sysml.runtime.matrix.operators.Operator;\n+import org.apache.sysml.runtime.matrix.operators.ReorgOperator;\n/**\n* Cpmm: cross-product matrix multiplication operation (distributed matrix multiply\n@@ -93,6 +99,23 @@ public class CpmmSPInstruction extends BinarySPInstruction {\nin2 = in2.filter(new FilterNonEmptyBlocksFunction());\n}\n+ if( SparkUtils.isHashPartitioned(in1) //ZIPMM-like CPMM\n+ && mc1.getNumRowBlocks()==1 && mc2.getCols()==1 ) {\n+ //note: if the major input is hash-partitioned and it's a matrix-vector\n+ //multiply, avoid the index mapping to preserve the partitioning similar\n+ //to a ZIPMM but with different transpose characteristics\n+ JavaRDD<MatrixBlock> out = in1\n+ .join(in2.mapToPair(new ReorgMapFunction(\"r'\")))\n+ .values().map(new Cpmm2MultiplyFunction())\n+ .filter(new FilterNonEmptyBlocksFunction2());\n+ MatrixBlock out2 = RDDAggregateUtils.sumStable(out);\n+\n+ //put output block into symbol table (no lineage because single block)\n+ //this also includes implicit maintenance of matrix characteristics\n+ sec.setMatrixOutput(output.getName(), out2, getExtendedOpcode());\n+ }\n+ else //GENERAL CPMM\n+ {\n//compute preferred join degree of parallelism\nint numPreferred = getPreferredParJoin(mc1, mc2, in1.getNumPartitions(), in2.getNumPartitions());\nint numPartJoin = Math.min(getMaxParJoin(mc1, mc2), numPreferred);\n@@ -128,6 +151,7 @@ public class CpmmSPInstruction extends BinarySPInstruction {\nupdateBinaryMMOutputMatrixCharacteristics(sec, true);\n}\n}\n+ }\nprivate static int getPreferredParJoin(MatrixCharacteristics mc1, MatrixCharacteristics mc2, int numPar1, int numPar2) {\nint defPar = SparkExecutionContext.getDefaultParallelism(true);\n@@ -190,4 +214,27 @@ public class CpmmSPInstruction extends BinarySPInstruction {\nreturn new Tuple2<>( ixOut, blkOut );\n}\n}\n+\n+ private static class Cpmm2MultiplyFunction implements Function<Tuple2<MatrixBlock,MatrixBlock>, MatrixBlock>\n+ {\n+ private static final long serialVersionUID = -3718880362385713416L;\n+ private AggregateBinaryOperator _op = null;\n+ private ReorgOperator _rop = null;\n+\n+ @Override\n+ public MatrixBlock call(Tuple2<MatrixBlock, MatrixBlock> arg0) throws Exception {\n+ //lazy operator construction\n+ if( _op == null ) {\n+ AggregateOperator agg = new AggregateOperator(0, Plus.getPlusFnObject());\n+ _op = new AggregateBinaryOperator(Multiply.getMultiplyFnObject(), agg);\n+ _rop = new ReorgOperator(SwapIndex.getSwapIndexFnObject());\n+ }\n+ //prepare inputs, including transpose of right-hand-side\n+ MatrixBlock in1 = arg0._1();\n+ MatrixBlock in2 = (MatrixBlock)arg0._2()\n+ .reorgOperations(_rop, new MatrixBlock(), 0, 0, 0);\n+ //core block matrix multiplication\n+ return in1.aggregateBinaryOperations(in1, in2, new MatrixBlock(), _op);\n+ }\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/FilterNonEmptyBlocksFunction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/FilterNonEmptyBlocksFunction.java", "diff": "@@ -28,20 +28,14 @@ import org.apache.sysml.runtime.matrix.data.MatrixIndexes;\npublic class FilterNonEmptyBlocksFunction implements Function<Tuple2<MatrixIndexes,MatrixBlock>, Boolean>\n{\n-\nprivate static final long serialVersionUID = -8856829325565589854L;\n@Override\n- public Boolean call(Tuple2<MatrixIndexes, MatrixBlock> arg0)\n- throws Exception\n- {\n+ public Boolean call(Tuple2<MatrixIndexes, MatrixBlock> arg0) throws Exception {\n//always keep 1-1 block in order to prevent empty rdds\nboolean ix1 = (arg0._1().getRowIndex()==1\n&& arg0._1().getColumnIndex()==1);\n-\n//returns true for non-empty matrix blocks\nreturn !arg0._2().isEmptyBlock(false) || ix1;\n}\n-\n-\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/functions/FilterNonEmptyBlocksFunction2.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.runtime.instructions.spark.functions;\n+\n+import org.apache.spark.api.java.function.Function;\n+\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+\n+public class FilterNonEmptyBlocksFunction2 implements Function<MatrixBlock, Boolean>\n+{\n+ private static final long serialVersionUID = -8435900761521598692L;\n+\n+ @Override\n+ public Boolean call(MatrixBlock arg0) throws Exception {\n+ return !arg0.isEmptyBlock(false);\n+ }\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/applications/L2SVMTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/applications/L2SVMTest.java", "diff": "@@ -33,7 +33,6 @@ import org.apache.sysml.test.utils.TestUtils;\npublic abstract class L2SVMTest extends AutomatedTestBase\n{\n-\nprotected final static String TEST_DIR = \"applications/l2svm/\";\nprotected final static String TEST_NAME = \"L2SVM\";\nprotected String TEST_CLASS_DIR = TEST_DIR + L2SVMTest.class.getSimpleName() + \"/\";\n@@ -51,12 +50,11 @@ public abstract class L2SVMTest extends AutomatedTestBase\n@Parameters\npublic static Collection<Object[]> data() {\n- Object[][] data = new Object[][] {\n+ return Arrays.asList(new Object[][] {\n//sparse tests (sparsity=0.01)\n{100, 50, 0.01, false}, {1000, 500, 0.01, false}, {10000, 750, 0.01, false}, {10000, 750, 0.01, true}, {100000, 1000, 0.01, false},\n//dense tests (sparsity=0.7)\n- {100, 50, 0.7, false}, {1000, 500, 0.7, false}, {1000, 500, 0.7, true}, {10000, 750, 0.7, false} };\n- return Arrays.asList(data);\n+ {100, 50, 0.7, false}, {1000, 500, 0.7, false}, {1000, 500, 0.7, true}, {10000, 750, 0.7, false} });\n}\n@Override\n@@ -66,13 +64,13 @@ public abstract class L2SVMTest extends AutomatedTestBase\nprotected void testL2SVM(ScriptType scriptType)\n{\n- System.out.println(\"------------ BEGIN \" + TEST_NAME + \" \" + scriptType + \" TEST WITH {\" + numRecords + \", \" + numFeatures\n+ System.out.println(\"------------ BEGIN \" + TEST_NAME + \" \" + scriptType\n+ + \" TEST WITH {\" + numRecords + \", \" + numFeatures\n+ \", \" + sparsity + \", \" + intercept + \"} ------------\");\nthis.scriptType = scriptType;\n-\nint rows = numRecords;\nint cols = numFeatures;\n- double epsilon = 1.0e-8;\n+ double epsilon = 1e-10;\ndouble lambda = 1.0;\nint maxiterations = 3;\nint maxNumberOfMRJobs = 21;\n@@ -94,9 +92,7 @@ public abstract class L2SVMTest extends AutomatedTestBase\nproArgs.add(\"model=\" + output(\"w\"));\nproArgs.add(\"Log=\" + output(\"Log\"));\nprogramArgs = proArgs.toArray(new String[proArgs.size()]);\n-\nfullDMLScriptName = getScript();\n-\nrCmd = getRCmd(inputDir(), (intercept ? Integer.toString(1) : Integer.toString(0)), Double.toString(epsilon),\nDouble.toString(lambda), Integer.toString(maxiterations), expectedDir());\n@@ -107,13 +103,12 @@ public abstract class L2SVMTest extends AutomatedTestBase\nwriteInputMatrixWithMTD(\"X\", X, true);\nwriteInputMatrixWithMTD(\"Y\", Y, true);\n-\nrunTest(true, EXCEPTION_NOT_EXPECTED, null, maxNumberOfMRJobs);\nrunRScript(true);\nHashMap<CellIndex, Double> wR = readRMatrixFromFS(\"w\");\nHashMap<CellIndex, Double> wSYSTEMML= readDMLMatrixFromHDFS(\"w\");\n- TestUtils.compareMatrices(wR, wSYSTEMML, Math.pow(10, -12), \"wR\", \"wSYSTEMML\");\n+ TestUtils.compareMatrices(wR, wSYSTEMML, epsilon, \"wR\", \"wSYSTEMML\");\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "diff": "@@ -52,6 +52,7 @@ import org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.compress.CompressedMatrixBlock;\nimport org.apache.sysml.runtime.controlprogram.Program;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.caching.FrameObject;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\n@@ -553,14 +554,7 @@ public class SparkExecutionContext extends ExecutionContext\n//create coarse-grained partitioned broadcasts\nif (numParts > 1) {\n- for( int i=0; i<numParts; i++ ) {\n- int offset = i * numPerPart;\n- int numBlks = Math.min(numPerPart, pmb.getNumRowBlocks()*pmb.getNumColumnBlocks()-offset);\n- PartitionedBlock<MatrixBlock> tmp = pmb.createPartition(offset, numBlks, new MatrixBlock());\n- ret[i] = getSparkContext().broadcast(tmp);\n- if( !isLocalMaster() )\n- tmp.clearBlocks();\n- }\n+ Arrays.parallelSetAll(ret, i -> createPartitionedBroadcast(pmb, numPerPart, i));\n}\nelse { //single partition\nret[0] = getSparkContext().broadcast(pmb);\n@@ -622,14 +616,7 @@ public class SparkExecutionContext extends ExecutionContext\n//create coarse-grained partitioned broadcasts\nif (numParts > 1) {\n- for( int i=0; i<numParts; i++ ) {\n- int offset = i * numPerPart;\n- int numBlks = Math.min(numPerPart, pmb.getNumRowBlocks()*pmb.getNumColumnBlocks()-offset);\n- PartitionedBlock<FrameBlock> tmp = pmb.createPartition(offset, numBlks, new FrameBlock());\n- ret[i] = getSparkContext().broadcast(tmp);\n- if( !isLocalMaster() )\n- tmp.clearBlocks();\n- }\n+ Arrays.parallelSetAll(ret, i -> createPartitionedBroadcast(pmb, numPerPart, i));\n}\nelse { //single partition\nret[0] = getSparkContext().broadcast(pmb);\n@@ -652,6 +639,17 @@ public class SparkExecutionContext extends ExecutionContext\nreturn bret;\n}\n+ private Broadcast<PartitionedBlock<? extends CacheBlock>> createPartitionedBroadcast(\n+ PartitionedBlock<? extends CacheBlock> pmb, int numPerPart, int pos) {\n+ int offset = pos * numPerPart;\n+ int numBlks = Math.min(numPerPart, pmb.getNumRowBlocks() * pmb.getNumColumnBlocks() - offset);\n+ PartitionedBlock<? extends CacheBlock> tmp = pmb.createPartition(offset, numBlks);\n+ Broadcast<PartitionedBlock<? extends CacheBlock>> ret = getSparkContext().broadcast(tmp);\n+ if (!isLocalMaster())\n+ tmp.clearBlocks();\n+ return ret;\n+ }\n+\n/**\n* Keep the output rdd of spark rdd operations as meta data of matrix/frame\n* objects in the symbol table.\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/PartitionedBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/PartitionedBlock.java", "diff": "@@ -29,6 +29,7 @@ import java.io.ObjectOutput;\nimport java.io.ObjectOutputStream;\nimport java.lang.reflect.Constructor;\nimport java.util.ArrayList;\n+import java.util.Arrays;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\n@@ -76,18 +77,16 @@ public class PartitionedBlock<T extends CacheBlock> implements Externalizable\nint ncblks = getNumColumnBlocks();\nint code = CacheBlockFactory.getCode(block);\n- try\n- {\n+ try {\n_partBlocks = new CacheBlock[nrblks * ncblks];\n- for( int i=0, ix=0; i<nrblks; i++ )\n- for( int j=0; j<ncblks; j++, ix++ ) {\n+ Arrays.parallelSetAll(_partBlocks, index -> {\n+ int i = index % nrblks;\n+ int j = index % ncblks;\nT tmp = (T) CacheBlockFactory.newInstance(code);\n- block.slice(i*_brlen, Math.min((i+1)*_brlen, rlen)-1,\n+ return block.slice(i * _brlen, Math.min((i + 1) * _brlen, rlen) - 1,\nj * _bclen, Math.min((j + 1) * _bclen, clen) - 1, tmp);\n- _partBlocks[ix] = tmp;\n- }\n- }\n- catch(Exception ex) {\n+ });\n+ } catch(Exception ex) {\nthrow new RuntimeException(\"Failed partitioning of broadcast variable input.\", ex);\n}\n@@ -107,7 +106,7 @@ public class PartitionedBlock<T extends CacheBlock> implements Externalizable\n_partBlocks = new CacheBlock[nrblks * ncblks];\n}\n- public PartitionedBlock<T> createPartition( int offset, int numBlks, T block )\n+ public PartitionedBlock<T> createPartition( int offset, int numBlks)\n{\nPartitionedBlock<T> ret = new PartitionedBlock<>();\nret._rlen = _rlen;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/PartitionedBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/PartitionedBlock.java", "diff": "@@ -80,7 +80,7 @@ public class PartitionedBlock<T extends CacheBlock> implements Externalizable\ntry {\n_partBlocks = new CacheBlock[nrblks * ncblks];\nArrays.parallelSetAll(_partBlocks, index -> {\n- int i = index % nrblks;\n+ int i = index / ncblks;\nint j = index % ncblks;\nT tmp = (T) CacheBlockFactory.newInstance(code);\nreturn block.slice(i * _brlen, Math.min((i + 1) * _brlen, rlen) - 1,\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "diff": "@@ -33,6 +33,7 @@ import org.apache.sysml.conf.DMLConfig;\nimport org.apache.sysml.hops.Hop.DataOpTypes;\nimport org.apache.sysml.hops.Hop.FileFormatTypes;\nimport org.apache.sysml.hops.Hop.OpOp2;\n+import org.apache.sysml.hops.Hop.ReOrgOp;\nimport org.apache.sysml.hops.rewrite.HopRewriteUtils;\nimport org.apache.sysml.lops.Checkpoint;\nimport org.apache.sysml.lops.Lop;\n@@ -924,10 +925,10 @@ public class OptimizerUtils\np.optFindExecType(); //ensure exec type evaluated\nret &= ( p.getExecType()==ExecType.CP\n||(p instanceof AggBinaryOp && allowsToFilterEmptyBlockOutputs(p) )\n- ||(p instanceof DataOp && ((DataOp)p).getDataOpType()==DataOpTypes.PERSISTENTWRITE && ((DataOp)p).getInputFormatType()==FileFormatTypes.TEXT))\n+ ||(HopRewriteUtils.isReorg(p, ReOrgOp.RESHAPE, ReOrgOp.TRANS) && allowsToFilterEmptyBlockOutputs(p) )\n+ ||(HopRewriteUtils.isData(p, DataOpTypes.PERSISTENTWRITE) && ((DataOp)p).getInputFormatType()==FileFormatTypes.TEXT))\n&& !(p instanceof FunctionOp || (p instanceof DataOp && ((DataOp)p).getInputFormatType()!=FileFormatTypes.TEXT) ); //no function call or transient write\n}\n-\nreturn ret;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/LocalParWorker.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/LocalParWorker.java", "diff": "@@ -138,7 +138,7 @@ public class LocalParWorker extends ParWorker implements Runnable\n}\nfinally {\n//cleanup fair scheduler pool for worker thread\n- if( OptimizerUtils.isSparkExecutionMode() ) {\n+ if( OptimizerUtils.isSparkExecutionMode() && pool != -1 ) {\nSparkExecutionContext sec = (SparkExecutionContext)_ec;\nsec.cleanupThreadLocalSchedulerPool(pool);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "diff": "@@ -43,6 +43,7 @@ import org.apache.sysml.runtime.instructions.spark.utils.SparkUtils;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\n+import org.apache.sysml.runtime.matrix.data.OperationsOnMatrixValues;\nimport org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;\nimport org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\n@@ -203,10 +204,10 @@ public class CpmmSPInstruction extends BinarySPInstruction {\nMatrixBlock blkIn1 = (MatrixBlock)arg0._2()._1().getValue();\nMatrixBlock blkIn2 = (MatrixBlock)arg0._2()._2().getValue();\nMatrixIndexes ixOut = new MatrixIndexes();\n- MatrixBlock blkOut = new MatrixBlock();\n//core block matrix multiplication\n- blkIn1.aggregateBinaryOperations(blkIn1, blkIn2, blkOut, _op);\n+ MatrixBlock blkOut = OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(blkIn1, blkIn2, new MatrixBlock(), _op);\n//return target block\nixOut.setIndexes(arg0._2()._1().getIndexes().getRowIndex(),\n@@ -234,7 +235,8 @@ public class CpmmSPInstruction extends BinarySPInstruction {\nMatrixBlock in2 = (MatrixBlock)arg0._2()\n.reorgOperations(_rop, new MatrixBlock(), 0, 0, 0);\n//core block matrix multiplication\n- return in1.aggregateBinaryOperations(in1, in2, new MatrixBlock(), _op);\n+ return OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(in1, in2, new MatrixBlock(), _op);\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "diff": "@@ -327,7 +327,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = _pbc.getBlock(1, (int)ixIn.getRowIndex());\n//execute matrix-vector mult\n- return (MatrixBlock) OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\n+ return OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\nleft, blkIn, new MatrixBlock(), _op);\n}\nelse //if( _type == CacheType.RIGHT )\n@@ -336,7 +336,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), 1);\n//execute matrix-vector mult\n- return (MatrixBlock) OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\n+ return OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\nblkIn, right, new MatrixBlock(), _op);\n}\n}\n@@ -392,7 +392,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = _pbc.getBlock(1, (int)ixIn.getRowIndex());\n//execute index preserving matrix multiplication\n- left.aggregateBinaryOperations(left, blkIn, blkOut, _op);\n+ OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(left, blkIn, blkOut, _op);\n}\nelse //if( _type == CacheType.RIGHT )\n{\n@@ -400,7 +400,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), 1);\n//execute index preserving matrix multiplication\n- blkIn.aggregateBinaryOperations(blkIn, right, blkOut, _op);\n+ OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(blkIn, right, blkOut, _op);\n}\nreturn new Tuple2<>(ixIn, blkOut);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/RmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/RmmSPInstruction.java", "diff": "@@ -43,6 +43,7 @@ import org.apache.sysml.runtime.instructions.spark.utils.SparkUtils;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\n+import org.apache.sysml.runtime.matrix.data.OperationsOnMatrixValues;\nimport org.apache.sysml.runtime.matrix.data.TripleIndexes;\nimport org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\n@@ -188,10 +189,10 @@ public class RmmSPInstruction extends BinarySPInstruction {\nMatrixIndexes ixOut = new MatrixIndexes(ixIn.getFirstIndex(), ixIn.getSecondIndex()); //i,j\nMatrixBlock blkIn1 = arg0._2()._1();\nMatrixBlock blkIn2 = arg0._2()._2();\n- MatrixBlock blkOut = new MatrixBlock();\n//core block matrix multiplication\n- blkIn1.aggregateBinaryOperations(blkIn1, blkIn2, blkOut, _op);\n+ MatrixBlock blkOut = OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(blkIn1, blkIn2, new MatrixBlock(), _op);\n//output new tuple\nreturn new Tuple2<>(ixOut, blkOut);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/Tsmm2SPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/Tsmm2SPInstruction.java", "diff": "@@ -215,7 +215,7 @@ public class Tsmm2SPInstruction extends UnarySPInstruction {\n(int)(_type.isLeft()?1:ixin.getColumnIndex()));\nMatrixBlock mbin2t = transpose(mbin2, new MatrixBlock()); //prep for transpose rewrite mm\n- MatrixBlock out2 = (MatrixBlock) OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes( //mm\n+ MatrixBlock out2 = OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes( //mm\n_type.isLeft() ? mbin2t : mbin, _type.isLeft() ? mbin : mbin2t, new MatrixBlock(), _op);\nMatrixIndexes ixout2 = _type.isLeft() ? new MatrixIndexes(2,1) : new MatrixIndexes(1,2);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ZipmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ZipmmSPInstruction.java", "diff": "@@ -36,6 +36,7 @@ import org.apache.sysml.runtime.instructions.cp.CPOperand;\nimport org.apache.sysml.runtime.instructions.spark.utils.RDDAggregateUtils;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\n+import org.apache.sysml.runtime.matrix.data.OperationsOnMatrixValues;\nimport org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\nimport org.apache.sysml.runtime.matrix.operators.Operator;\n@@ -124,7 +125,8 @@ public class ZipmmSPInstruction extends BinarySPInstruction {\nMatrixBlock tmp = (MatrixBlock)in2.reorgOperations(_rop, new MatrixBlock(), 0, 0, 0);\n//core matrix multiplication (for t(y)%*%X or t(X)%*%y)\n- return tmp.aggregateBinaryOperations(tmp, in1, new MatrixBlock(), _abop);\n+ return OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(tmp, in1, new MatrixBlock(), _abop);\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -228,14 +228,13 @@ public class OperationsOnMatrixValues\nvalue1.aggregateBinaryOperations(indexes1, value1, indexes2, value2, valueOut, op);\n}\n- public static MatrixValue performAggregateBinaryIgnoreIndexes(MatrixBlock value1, MatrixBlock value2,\n+ public static MatrixBlock performAggregateBinaryIgnoreIndexes(MatrixBlock value1, MatrixBlock value2,\nMatrixBlock valueOut, AggregateBinaryOperator op) {\n//perform on the value\nif( value2 instanceof CompressedMatrixBlock )\n- value2.aggregateBinaryOperations(value1, value2, valueOut, op);\n+ return value2.aggregateBinaryOperations(value1, value2, valueOut, op);\nelse\n- value1.aggregateBinaryOperations(value1, value2, valueOut, op);\n- return valueOut;\n+ return value1.aggregateBinaryOperations(value1, value2, valueOut, op);\n}\n@SuppressWarnings(\"rawtypes\")\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/DataGenOp.java", "new_path": "src/main/java/org/apache/sysml/hops/DataGenOp.java", "diff": "@@ -374,8 +374,10 @@ public class DataGenOp extends Hop implements MultiThreadedHop\nreturn getInput().get(getParamIndex(key));\n}\n- public void setInput(String key, Hop hop) {\n+ public void setInput(String key, Hop hop, boolean linkParent) {\ngetInput().set(getParamIndex(key), hop);\n+ if( linkParent )\n+ hop.getParent().add(this);\n}\npublic boolean hasConstantValue()\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "diff": "@@ -438,8 +438,8 @@ public class RewriteAlgebraicSimplificationStatic extends HopRewriteRule\n&& HopRewriteUtils.isLiteralOfValue(min, 0)\n&& HopRewriteUtils.isLiteralOfValue(max, 0) )\n{\n- gen.setInput(DataExpression.RAND_MIN, right);\n- gen.setInput(DataExpression.RAND_MAX, right);\n+ gen.setInput(DataExpression.RAND_MIN, right, true);\n+ gen.setInput(DataExpression.RAND_MAX, right, true);\n//rewire all parents (avoid anomalies with replicated datagen)\nList<Hop> parents = new ArrayList<>(bop.getParent());\nfor( Hop p : parents )\n@@ -454,8 +454,8 @@ public class RewriteAlgebraicSimplificationStatic extends HopRewriteRule\n&& HopRewriteUtils.isLiteralOfValue(max, 1) )\n{\nif( HopRewriteUtils.isLiteralOfValue(min, 1) )\n- gen.setInput(DataExpression.RAND_MIN, right);\n- gen.setInput(DataExpression.RAND_MAX, right);\n+ gen.setInput(DataExpression.RAND_MIN, right, true);\n+ gen.setInput(DataExpression.RAND_MAX, right, true);\n//rewire all parents (avoid anomalies with replicated datagen)\nList<Hop> parents = new ArrayList<>(bop.getParent());\nfor( Hop p : parents )\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "diff": "@@ -431,43 +431,27 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nthrows Exception\n{\nArrayList<Tuple2<MatrixIndexes, MatrixBlock>> ret = new ArrayList<>();\n-\nMatrixIndexes ixIn = arg0._1();\nMatrixBlock blkIn = arg0._2();\n- if( _type == CacheType.LEFT )\n- {\n+ if( _type == CacheType.LEFT ) {\n//for all matching left-hand-side blocks\nint len = _pbc.getNumRowBlocks();\n- for( int i=1; i<=len; i++ )\n- {\n+ for( int i=1; i<=len; i++ ) {\nMatrixBlock left = _pbc.getBlock(i, (int)ixIn.getRowIndex());\n- MatrixIndexes ixOut = new MatrixIndexes();\n- MatrixBlock blkOut = new MatrixBlock();\n-\n- //execute matrix-vector mult\n- OperationsOnMatrixValues.performAggregateBinary(\n- new MatrixIndexes(i,ixIn.getRowIndex()), left, ixIn, blkIn, ixOut, blkOut, _op);\n-\n- ret.add(new Tuple2<>(ixOut, blkOut));\n+ MatrixBlock blkOut = OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(left, blkIn, new MatrixBlock(), _op);\n+ ret.add(new Tuple2<>(new MatrixIndexes(i, ixIn.getColumnIndex()), blkOut));\n}\n}\n- else //if( _type == CacheType.RIGHT )\n- {\n+ else { //RIGHT\n//for all matching right-hand-side blocks\nint len = _pbc.getNumColumnBlocks();\n- for( int j=1; j<=len; j++ )\n- {\n- //get the right hand side matrix\n+ for( int j=1; j<=len; j++ ) {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), j);\n- MatrixIndexes ixOut = new MatrixIndexes();\n- MatrixBlock blkOut = new MatrixBlock();\n-\n- //execute matrix-vector mult\n- OperationsOnMatrixValues.performAggregateBinary(\n- ixIn, blkIn, new MatrixIndexes(ixIn.getColumnIndex(),j), right, ixOut, blkOut, _op);\n-\n- ret.add(new Tuple2<>(ixOut, blkOut));\n+ MatrixBlock blkOut = OperationsOnMatrixValues\n+ .performAggregateBinaryIgnoreIndexes(blkIn, right, new MatrixBlock(), _op);\n+ ret.add(new Tuple2<>(new MatrixIndexes(ixIn.getRowIndex(), j), blkOut));\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -1490,7 +1490,8 @@ public class LibMatrixMult\n*/\nprivate static void matrixMultUltraSparse(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret, int rl, int ru) {\nfinal boolean leftUS = m1.isUltraSparse()\n- || (m1.isUltraSparse(false) && !m2.isUltraSparse());\n+ || (m1.isUltraSparse(false) && !m2.isUltraSparse())\n+ || (m1.sparse && !m2.sparse);\nfinal int m = m1.rlen;\nfinal int cd = m1.clen;\nfinal int n = m2.clen;\n@@ -3693,14 +3694,21 @@ public class LibMatrixMult\n}\npublic static boolean isUltraSparseMatrixMult(MatrixBlock m1, MatrixBlock m2) {\n+ if( m2.clen == 1 ) //mv always dense\n+ return false;\n//note: ultra-sparse matrix mult implies also sparse outputs, hence we need\n//to be conservative an cannot use this for all ultra-sparse matrices.\n+ double outSp = OptimizerUtils.getMatMultSparsity(\n+ m1.getSparsity(), m2.getSparsity(), m1.rlen, m1.clen, m2.clen, true);\nreturn (m1.isUltraSparse() || m2.isUltraSparse()) //base case\n|| (m1.isUltraSparsePermutationMatrix()\n&& OptimizerUtils.getSparsity(m2.rlen, m2.clen, m2.nonZeros)<1.0)\n|| ((m1.isUltraSparse(false) || m2.isUltraSparse(false))\n- && OptimizerUtils.getMatMultSparsity(m1.getSparsity(), m2.getSparsity(),\n- m1.rlen, m1.clen, m2.clen, true) < MatrixBlock.ULTRA_SPARSITY_TURN_POINT2);\n+ && outSp < MatrixBlock.ULTRA_SPARSITY_TURN_POINT2)\n+ || (m1.getSparsity() < MatrixBlock.ULTRA_SPARSITY_TURN_POINT2\n+ && m1.getNonZeros() < MatrixBlock.ULTRA_SPARSE_BLOCK_NNZ\n+ && m1.getLength()+m2.getLength() < (long)m1.rlen*m2.clen\n+ && outSp < MatrixBlock.SPARSITY_TURN_POINT);\n}\nprivate static MatrixBlock prepMatrixMultRightInput( MatrixBlock m1, MatrixBlock m2 ) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -101,6 +101,7 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\n//sparsity threshold for ultra-sparse matrix operations (40nnz in a 1kx1k block)\npublic static final double ULTRA_SPARSITY_TURN_POINT = 0.00004;\npublic static final double ULTRA_SPARSITY_TURN_POINT2 = 0.0004;\n+ public static final int ULTRA_SPARSE_BLOCK_NNZ = 40;\n//default sparse block type: modified compressed sparse rows, for efficient incremental construction\npublic static final SparseBlock.Type DEFAULT_SPARSEBLOCK = SparseBlock.Type.MCSR;\n//default sparse block type for update in place: compressed sparse rows, to prevent serialization\n@@ -874,7 +875,8 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\npublic boolean isUltraSparse(boolean checkNnz) {\ndouble sp = ((double)nonZeros/rlen)/clen;\n//check for sparse representation in order to account for vectors in dense\n- return sparse && sp<ULTRA_SPARSITY_TURN_POINT && (!checkNnz || nonZeros<40);\n+ return sparse && sp<ULTRA_SPARSITY_TURN_POINT\n+ && (!checkNnz || nonZeros<ULTRA_SPARSE_BLOCK_NNZ);\n}\npublic boolean isUltraSparsePermutationMatrix() {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -217,15 +217,15 @@ public class OperationsOnMatrixValues\nvalueIn.aggregateUnaryOperations(op, valueOut, brlen, bclen, indexesIn);\n}\n- public static void performAggregateBinary(MatrixIndexes indexes1, MatrixBlock value1, MatrixIndexes indexes2, MatrixBlock value2,\n+ public static MatrixBlock performAggregateBinary(MatrixIndexes indexes1, MatrixBlock value1, MatrixIndexes indexes2, MatrixBlock value2,\nMatrixIndexes indexesOut, MatrixBlock valueOut, AggregateBinaryOperator op) {\n//compute output index\nindexesOut.setIndexes(indexes1.getRowIndex(), indexes2.getColumnIndex());\n//perform on the value\nif( value2 instanceof CompressedMatrixBlock )\n- value2.aggregateBinaryOperations(value1, value2, valueOut, op);\n+ return value2.aggregateBinaryOperations(value1, value2, valueOut, op);\nelse //default\n- value1.aggregateBinaryOperations(indexes1, value1, indexes2, value2, valueOut, op);\n+ return value1.aggregateBinaryOperations(indexes1, value1, indexes2, value2, valueOut, op);\n}\npublic static MatrixBlock performAggregateBinaryIgnoreIndexes(MatrixBlock value1, MatrixBlock value2,\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteCommonSubexpressionElimination.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteCommonSubexpressionElimination.java", "diff": "@@ -25,6 +25,7 @@ import java.util.HashMap;\nimport org.apache.sysml.hops.DataOp;\nimport org.apache.sysml.hops.Hop;\nimport org.apache.sysml.hops.LiteralOp;\n+import org.apache.sysml.hops.Hop.DataOpTypes;\n/**\n* Rule: CommonSubexpressionElimination. For all statement blocks,\n@@ -99,7 +100,8 @@ public class RewriteCommonSubexpressionElimination extends HopRewriteRule\nif( hop.isVisited() )\nreturn ret;\n- if( hop.getInput().isEmpty() ) //LEAF NODE\n+ if( hop.getInput().isEmpty() //LEAF NODE\n+ || HopRewriteUtils.isData(hop, DataOpTypes.TRANSIENTREAD) )\n{\nif( hop instanceof LiteralOp )\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/SparkUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/SparkUtils.java", "diff": "@@ -70,40 +70,28 @@ public class SparkUtils\nreturn new Tuple2<>(in.getIndexes(), (MatrixBlock)in.getValue());\n}\n- public static ArrayList<Tuple2<MatrixIndexes,MatrixBlock>> fromIndexedMatrixBlock( ArrayList<IndexedMatrixValue> in ) {\n- ArrayList<Tuple2<MatrixIndexes,MatrixBlock>> ret = new ArrayList<>();\n- for( IndexedMatrixValue imv : in )\n- ret.add(fromIndexedMatrixBlock(imv));\n- return ret;\n+ public static List<Tuple2<MatrixIndexes,MatrixBlock>> fromIndexedMatrixBlock( List<IndexedMatrixValue> in ) {\n+ return in.stream().map(imv -> fromIndexedMatrixBlock(imv)).collect(Collectors.toList());\n}\npublic static Pair<MatrixIndexes,MatrixBlock> fromIndexedMatrixBlockToPair( IndexedMatrixValue in ){\nreturn new Pair<>(in.getIndexes(), (MatrixBlock)in.getValue());\n}\n- public static ArrayList<Pair<MatrixIndexes,MatrixBlock>> fromIndexedMatrixBlockToPair( ArrayList<IndexedMatrixValue> in ) {\n- ArrayList<Pair<MatrixIndexes,MatrixBlock>> ret = new ArrayList<>();\n- for( IndexedMatrixValue imv : in )\n- ret.add(fromIndexedMatrixBlockToPair(imv));\n- return ret;\n+ public static List<Pair<MatrixIndexes,MatrixBlock>> fromIndexedMatrixBlockToPair( List<IndexedMatrixValue> in ) {\n+ return in.stream().map(imv -> fromIndexedMatrixBlockToPair(imv)).collect(Collectors.toList());\n}\npublic static Tuple2<Long,FrameBlock> fromIndexedFrameBlock( Pair<Long, FrameBlock> in ){\nreturn new Tuple2<>(in.getKey(), in.getValue());\n}\n- public static ArrayList<Tuple2<Long,FrameBlock>> fromIndexedFrameBlock( ArrayList<Pair<Long, FrameBlock>> in ) {\n- ArrayList<Tuple2<Long, FrameBlock>> ret = new ArrayList<>();\n- for( Pair<Long, FrameBlock> ifv : in )\n- ret.add(fromIndexedFrameBlock(ifv));\n- return ret;\n+ public static List<Tuple2<Long,FrameBlock>> fromIndexedFrameBlock(List<Pair<Long, FrameBlock>> in) {\n+ return in.stream().map(ifv -> fromIndexedFrameBlock(ifv)).collect(Collectors.toList());\n}\n- public static ArrayList<Pair<Long,Long>> toIndexedLong( List<Tuple2<Long, Long>> in ) {\n- ArrayList<Pair<Long, Long>> ret = new ArrayList<>();\n- for( Tuple2<Long, Long> e : in )\n- ret.add(new Pair<>(e._1(), e._2()));\n- return ret;\n+ public static List<Pair<Long,Long>> toIndexedLong( List<Tuple2<Long, Long>> in ) {\n+ return in.stream().map(e -> new Pair<>(e._1(), e._2())).collect(Collectors.toList());\n}\npublic static Pair<Long,FrameBlock> toIndexedFrameBlock( Tuple2<Long,FrameBlock> in ) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -22,6 +22,7 @@ package org.apache.sysml.runtime.matrix.data;\nimport java.util.ArrayList;\nimport java.util.Arrays;\n+import java.util.List;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n@@ -238,7 +239,7 @@ public class OperationsOnMatrixValues\n}\n@SuppressWarnings(\"rawtypes\")\n- public static ArrayList performSlice(IndexRange ixrange, int brlen, int bclen, int iix, int jix, CacheBlock in) {\n+ public static List performSlice(IndexRange ixrange, int brlen, int bclen, int iix, int jix, CacheBlock in) {\nif( in instanceof MatrixBlock )\nreturn performSlice(ixrange, brlen, bclen, iix, jix, (MatrixBlock)in);\nelse if( in instanceof FrameBlock )\n@@ -246,13 +247,11 @@ public class OperationsOnMatrixValues\nthrow new DMLRuntimeException(\"Unsupported cache block type: \"+in.getClass().getName());\n}\n-\n@SuppressWarnings(\"rawtypes\")\n- public static ArrayList performSlice(IndexRange ixrange, int brlen, int bclen, int iix, int jix, MatrixBlock in) {\n+ public static List performSlice(IndexRange ixrange, int brlen, int bclen, int iix, int jix, MatrixBlock in) {\nIndexedMatrixValue imv = new IndexedMatrixValue(new MatrixIndexes(iix, jix), (MatrixBlock)in);\nArrayList<IndexedMatrixValue> outlist = new ArrayList<>();\nperformSlice(imv, ixrange, brlen, bclen, outlist);\n-\nreturn SparkUtils.fromIndexedMatrixBlockToPair(outlist);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseRowVector.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/SparseRowVector.java", "diff": "@@ -209,8 +209,9 @@ public final class SparseRowVector extends SparseRow implements Serializable\nreturn (index >= 0) ? values[index] : 0;\n}\n- public int searchIndexesFirstLTE(int col)\n- {\n+ public int searchIndexesFirstLTE(int col) {\n+ if( size == 0 ) return -1;\n+\n//search for existing col index\nint index = Arrays.binarySearch(indexes, 0, size, col);\nif( index >= 0 )\n@@ -221,8 +222,9 @@ public final class SparseRowVector extends SparseRow implements Serializable\nreturn (index-1 < size) ? index-1 : -1;\n}\n- public int searchIndexesFirstGTE(int col)\n- {\n+ public int searchIndexesFirstGTE(int col) {\n+ if( size == 0 ) return -1;\n+\n//search for existing col index\nint index = Arrays.binarySearch(indexes, 0, size, col);\nif( index >= 0 )\n@@ -233,8 +235,9 @@ public final class SparseRowVector extends SparseRow implements Serializable\nreturn (index < size) ? index : -1;\n}\n- public int searchIndexesFirstGT(int col)\n- {\n+ public int searchIndexesFirstGT(int col) {\n+ if( size == 0 ) return -1;\n+\n//search for existing col index\nint index = Arrays.binarySearch(indexes, 0, size, col);\nif( index >= 0 )\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/mr/AggregateBinaryInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/mr/AggregateBinaryInstruction.java", "diff": "@@ -154,7 +154,7 @@ public class AggregateBinaryInstruction extends BinaryMRInstructionBase implemen\nout=cachedValues.holdPlace(output, valueClass);\n//process instruction\n- OperationsOnMatrixValues.performAggregateBinary(\n+ OperationsOnMatrixValues.matMult(\nin1.getIndexes(), (MatrixBlock) in1.getValue(),\nin2.getIndexes(), (MatrixBlock) in2.getValue(),\nout.getIndexes(), (MatrixBlock) out.getValue(),\n@@ -200,7 +200,7 @@ public class AggregateBinaryInstruction extends BinaryMRInstructionBase implemen\nIndexedMatrixValue out = cachedValues.holdPlace(output, valueClass);\n//process instruction\n- OperationsOnMatrixValues.performAggregateBinary(in1.getIndexes(), (MatrixBlock)in1.getValue(),\n+ OperationsOnMatrixValues.matMult(in1.getIndexes(), (MatrixBlock)in1.getValue(),\nin2BlockIndex, (MatrixBlock) in2BlockValue, out.getIndexes(), (MatrixBlock)out.getValue(),\n((AggregateBinaryOperator)optr));\nremoveOutput &= ( !_outputEmptyBlocks && out.getValue().isEmpty() );\n@@ -227,7 +227,7 @@ public class AggregateBinaryInstruction extends BinaryMRInstructionBase implemen\nIndexedMatrixValue out = cachedValues.holdPlace(output, valueClass);\n//process instruction\n- OperationsOnMatrixValues.performAggregateBinary(\n+ OperationsOnMatrixValues.matMult(\nin1BlockIndex, (MatrixBlock)in1BlockValue,\nin2.getIndexes(), (MatrixBlock)in2.getValue(),\nout.getIndexes(), (MatrixBlock)out.getValue(),\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/mr/MatrixReshapeMRInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/mr/MatrixReshapeMRInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.mr;\nimport java.util.ArrayList;\n+import java.util.List;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.instructions.InstructionUtils;\n@@ -36,8 +37,6 @@ public class MatrixReshapeMRInstruction extends UnaryInstruction {\nprivate MatrixCharacteristics _mcIn = null;\nprivate MatrixCharacteristics _mcOut = null;\n- private ArrayList<IndexedMatrixValue> _cache = null;\n-\nprivate MatrixReshapeMRInstruction(Operator op, byte in, long rows, long cols, boolean byrow, byte out,\nString istr) {\nsuper(MRType.MMTSJ, op, in, out, istr);\n@@ -46,8 +45,7 @@ public class MatrixReshapeMRInstruction extends UnaryInstruction {\n_byrow = byrow;\n}\n- public void setMatrixCharacteristics( MatrixCharacteristics mcIn, MatrixCharacteristics mcOut )\n- {\n+ public void setMatrixCharacteristics( MatrixCharacteristics mcIn, MatrixCharacteristics mcOut ) {\n_mcIn = mcIn;\n}\n@@ -73,25 +71,17 @@ public class MatrixReshapeMRInstruction extends UnaryInstruction {\n{\nArrayList<IndexedMatrixValue> blkList = cachedValues.get(input);\nif( blkList != null )\n- for(IndexedMatrixValue imv : blkList)\n- {\n- if( imv == null )\n- continue;\n-\n- //get cached blocks\n- ArrayList<IndexedMatrixValue> out = _cache;\n+ for(IndexedMatrixValue imv : blkList) {\n+ if( imv == null ) continue;\n//process instruction\n_mcOut.setBlockSize(brlen, bclen);\n- out = LibMatrixReorg.reshape(imv, _mcIn, out, _mcOut, _byrow, true);\n+ List<IndexedMatrixValue> out =\n+ LibMatrixReorg.reshape(imv, _mcIn, _mcOut, _byrow, true);\n//put the output values in the output cache\nfor( IndexedMatrixValue outBlk : out )\ncachedValues.add(output, outBlk);\n-\n- //put blocks into own cache\n- if( LibMatrixReorg.ALLOW_BLOCK_REUSE )\n- _cache = out;\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/CpmmSPInstruction.java", "diff": "@@ -207,7 +207,7 @@ public class CpmmSPInstruction extends BinarySPInstruction {\n//core block matrix multiplication\nMatrixBlock blkOut = OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(blkIn1, blkIn2, new MatrixBlock(), _op);\n+ .matMult(blkIn1, blkIn2, new MatrixBlock(), _op);\n//return target block\nixOut.setIndexes(arg0._2()._1().getIndexes().getRowIndex(),\n@@ -236,7 +236,7 @@ public class CpmmSPInstruction extends BinarySPInstruction {\n.reorgOperations(_rop, new MatrixBlock(), 0, 0, 0);\n//core block matrix multiplication\nreturn OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(in1, in2, new MatrixBlock(), _op);\n+ .matMult(in1, in2, new MatrixBlock(), _op);\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MapmmSPInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.spark;\n-import java.util.ArrayList;\nimport java.util.Iterator;\n+import java.util.stream.IntStream;\nimport org.apache.spark.api.java.JavaPairRDD;\nimport org.apache.spark.api.java.JavaRDD;\n@@ -275,8 +275,8 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = _pbc.getBlock(1, (int)ixIn.getRowIndex());\n//execute matrix-vector mult\n- OperationsOnMatrixValues.performAggregateBinary(\n- new MatrixIndexes(1,ixIn.getRowIndex()), left, ixIn, blkIn, ixOut, blkOut, _op);\n+ OperationsOnMatrixValues.matMult(new MatrixIndexes(1,ixIn.getRowIndex()),\n+ left, ixIn, blkIn, ixOut, blkOut, _op);\n}\nelse //if( _type == CacheType.RIGHT )\n{\n@@ -284,8 +284,8 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), 1);\n//execute matrix-vector mult\n- OperationsOnMatrixValues.performAggregateBinary(\n- ixIn, blkIn, new MatrixIndexes(ixIn.getColumnIndex(),1), right, ixOut, blkOut, _op);\n+ OperationsOnMatrixValues.matMult(ixIn, blkIn,\n+ new MatrixIndexes(ixIn.getColumnIndex(),1), right, ixOut, blkOut, _op);\n}\n//output new tuple\n@@ -327,7 +327,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = _pbc.getBlock(1, (int)ixIn.getRowIndex());\n//execute matrix-vector mult\n- return OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\n+ return OperationsOnMatrixValues.matMult(\nleft, blkIn, new MatrixBlock(), _op);\n}\nelse //if( _type == CacheType.RIGHT )\n@@ -336,7 +336,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), 1);\n//execute matrix-vector mult\n- return OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(\n+ return OperationsOnMatrixValues.matMult(\nblkIn, right, new MatrixBlock(), _op);\n}\n}\n@@ -392,7 +392,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = _pbc.getBlock(1, (int)ixIn.getRowIndex());\n//execute index preserving matrix multiplication\n- OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(left, blkIn, blkOut, _op);\n+ OperationsOnMatrixValues.matMult(left, blkIn, blkOut, _op);\n}\nelse //if( _type == CacheType.RIGHT )\n{\n@@ -400,7 +400,7 @@ public class MapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), 1);\n//execute index preserving matrix multiplication\n- OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes(blkIn, right, blkOut, _op);\n+ OperationsOnMatrixValues.matMult(blkIn, right, blkOut, _op);\n}\nreturn new Tuple2<>(ixIn, blkOut);\n@@ -430,32 +430,23 @@ public class MapmmSPInstruction extends BinarySPInstruction {\npublic Iterator<Tuple2<MatrixIndexes, MatrixBlock>> call( Tuple2<MatrixIndexes, MatrixBlock> arg0 )\nthrows Exception\n{\n- ArrayList<Tuple2<MatrixIndexes, MatrixBlock>> ret = new ArrayList<>();\nMatrixIndexes ixIn = arg0._1();\nMatrixBlock blkIn = arg0._2();\nif( _type == CacheType.LEFT ) {\n- //for all matching left-hand-side blocks\n- int len = _pbc.getNumRowBlocks();\n- for( int i=1; i<=len; i++ ) {\n- MatrixBlock left = _pbc.getBlock(i, (int)ixIn.getRowIndex());\n- MatrixBlock blkOut = OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(left, blkIn, new MatrixBlock(), _op);\n- ret.add(new Tuple2<>(new MatrixIndexes(i, ixIn.getColumnIndex()), blkOut));\n- }\n+ //for all matching left-hand-side blocks, returned as lazy iterator\n+ return IntStream.range(1, _pbc.getNumRowBlocks()+1).mapToObj(i ->\n+ new Tuple2<>(new MatrixIndexes(i, ixIn.getColumnIndex()),\n+ OperationsOnMatrixValues.matMult(_pbc.getBlock(i, (int)ixIn.getRowIndex()), blkIn,\n+ new MatrixBlock(), _op))).iterator();\n}\nelse { //RIGHT\n- //for all matching right-hand-side blocks\n- int len = _pbc.getNumColumnBlocks();\n- for( int j=1; j<=len; j++ ) {\n- MatrixBlock right = _pbc.getBlock((int)ixIn.getColumnIndex(), j);\n- MatrixBlock blkOut = OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(blkIn, right, new MatrixBlock(), _op);\n- ret.add(new Tuple2<>(new MatrixIndexes(ixIn.getRowIndex(), j), blkOut));\n- }\n+ //for all matching right-hand-side blocks, returned as lazy iterator\n+ return IntStream.range(1, _pbc.getNumColumnBlocks()+1).mapToObj(j ->\n+ new Tuple2<>(new MatrixIndexes(ixIn.getRowIndex(), j),\n+ OperationsOnMatrixValues.matMult(blkIn, _pbc.getBlock((int)ixIn.getColumnIndex(), j),\n+ new MatrixBlock(), _op))).iterator();\n}\n-\n- return ret.iterator();\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MatrixReshapeSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/MatrixReshapeSPInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.spark;\n-import java.util.ArrayList;\nimport java.util.Iterator;\n+import java.util.List;\nimport org.apache.spark.api.java.JavaPairRDD;\nimport org.apache.spark.api.java.function.PairFlatMapFunction;\n@@ -138,8 +138,8 @@ public class MatrixReshapeSPInstruction extends UnarySPInstruction\nIndexedMatrixValue in = SparkUtils.toIndexedMatrixBlock(arg0);\n//execute actual reshape operation\n- ArrayList<IndexedMatrixValue> out = LibMatrixReorg\n- .reshape(in, _mcIn, new ArrayList<>(), _mcOut, _byrow, _outputEmptyBlocks);\n+ List<IndexedMatrixValue> out = LibMatrixReorg\n+ .reshape(in, _mcIn, _mcOut, _byrow, _outputEmptyBlocks);\n//output conversion (for compatibility w/ rdd schema)\nreturn SparkUtils.fromIndexedMatrixBlock(out).iterator();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/PMapmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/PMapmmSPInstruction.java", "diff": "@@ -193,8 +193,8 @@ public class PMapmmSPInstruction extends BinarySPInstruction {\nMatrixBlock left = pm.getBlock(i, (int)ixIn.getRowIndex());\n//execute matrix-vector mult\n- OperationsOnMatrixValues.performAggregateBinary(\n- new MatrixIndexes(i,ixIn.getRowIndex()), left, ixIn, blkIn, ixOut, blkOut, _op);\n+ OperationsOnMatrixValues.matMult(new MatrixIndexes(i,ixIn.getRowIndex()),\n+ left, ixIn, blkIn, ixOut, blkOut, _op);\n//output new tuple\nixOut.setIndexes(_offset+i, ixOut.getColumnIndex());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/RmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/RmmSPInstruction.java", "diff": "@@ -191,8 +191,7 @@ public class RmmSPInstruction extends BinarySPInstruction {\nMatrixBlock blkIn2 = arg0._2()._2();\n//core block matrix multiplication\n- MatrixBlock blkOut = OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(blkIn1, blkIn2, new MatrixBlock(), _op);\n+ MatrixBlock blkOut = OperationsOnMatrixValues.matMult(blkIn1, blkIn2, new MatrixBlock(), _op);\n//output new tuple\nreturn new Tuple2<>(ixOut, blkOut);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/Tsmm2SPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/Tsmm2SPInstruction.java", "diff": "@@ -154,7 +154,7 @@ public class Tsmm2SPInstruction extends UnarySPInstruction {\n(int)(_type.isLeft()?1:ixin.getColumnIndex()));\nMatrixBlock mbin2t = transpose(mbin2, new MatrixBlock()); //prep for transpose rewrite mm\n- MatrixBlock out2 = (MatrixBlock) OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes( //mm\n+ MatrixBlock out2 = (MatrixBlock) OperationsOnMatrixValues.matMult( //mm\n_type.isLeft() ? mbin2t : mbin, _type.isLeft() ? mbin : mbin2t, new MatrixBlock(), _op);\nMatrixIndexes ixout2 = _type.isLeft() ? new MatrixIndexes(2,1) : new MatrixIndexes(1,2);\nret.add(new Tuple2<>(ixout2, out2));\n@@ -215,7 +215,7 @@ public class Tsmm2SPInstruction extends UnarySPInstruction {\n(int)(_type.isLeft()?1:ixin.getColumnIndex()));\nMatrixBlock mbin2t = transpose(mbin2, new MatrixBlock()); //prep for transpose rewrite mm\n- MatrixBlock out2 = OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes( //mm\n+ MatrixBlock out2 = OperationsOnMatrixValues.matMult( //mm\n_type.isLeft() ? mbin2t : mbin, _type.isLeft() ? mbin : mbin2t, new MatrixBlock(), _op);\nMatrixIndexes ixout2 = _type.isLeft() ? new MatrixIndexes(2,1) : new MatrixIndexes(1,2);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ZipmmSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ZipmmSPInstruction.java", "diff": "@@ -125,8 +125,7 @@ public class ZipmmSPInstruction extends BinarySPInstruction {\nMatrixBlock tmp = (MatrixBlock)in2.reorgOperations(_rop, new MatrixBlock(), 0, 0, 0);\n//core matrix multiplication (for t(y)%*%X or t(X)%*%y)\n- return OperationsOnMatrixValues\n- .performAggregateBinaryIgnoreIndexes(tmp, in1, new MatrixBlock(), _abop);\n+ return OperationsOnMatrixValues.matMult(tmp, in1, new MatrixBlock(), _abop);\n}\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "diff": "@@ -28,10 +28,11 @@ import java.util.HashMap;\nimport java.util.HashSet;\nimport java.util.Iterator;\nimport java.util.List;\n-import java.util.Map.Entry;\n+import java.util.Map;\nimport java.util.concurrent.Callable;\nimport java.util.concurrent.ExecutorService;\nimport java.util.concurrent.Future;\n+import java.util.stream.Collectors;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject.UpdateType;\n@@ -69,9 +70,6 @@ public class LibMatrixReorg\n//safe due to copy-on-write and safe update-in-place handling)\npublic static final boolean SHALLOW_COPY_REORG = true;\n- //allow reuse of temporary blocks for certain operations\n- public static final boolean ALLOW_BLOCK_REUSE = false;\n-\n//use csr instead of mcsr sparse block for rexpand columns / diag v2m\npublic static final boolean SPARSE_OUTPUTS_IN_CSR = true;\n@@ -465,15 +463,15 @@ public class LibMatrixReorg\n* @param outputEmptyBlocks output blocks with nnz=0\n* @return list of indexed matrix values\n*/\n- public static ArrayList<IndexedMatrixValue> reshape( IndexedMatrixValue in, MatrixCharacteristics mcIn,\n- ArrayList<IndexedMatrixValue> out, MatrixCharacteristics mcOut, boolean rowwise, boolean outputEmptyBlocks ) {\n+ public static List<IndexedMatrixValue> reshape(IndexedMatrixValue in, MatrixCharacteristics mcIn,\n+ MatrixCharacteristics mcOut, boolean rowwise, boolean outputEmptyBlocks ) {\n//prepare inputs\nMatrixIndexes ixIn = in.getIndexes();\nMatrixBlock mbIn = (MatrixBlock) in.getValue();\n//prepare result blocks (no reuse in order to guarantee mem constraints)\nCollection<MatrixIndexes> rix = computeAllResultBlockIndexes(ixIn, mcIn, mcOut, mbIn, rowwise, outputEmptyBlocks);\n- HashMap<MatrixIndexes, MatrixBlock> rblk = createAllResultBlocks(rix, mbIn.nonZeros, mcIn, mcOut, rowwise, out);\n+ Map<MatrixIndexes, MatrixBlock> rblk = createAllResultBlocks(rix, mbIn.nonZeros, mcOut);\n//basic algorithm\nlong row_offset = (ixIn.getRowIndex()-1)*mcIn.getRowsPerBlock();\n@@ -483,15 +481,11 @@ public class LibMatrixReorg\nelse //dense\nreshapeDense(mbIn, row_offset, col_offset, rblk, mcIn, mcOut, rowwise);\n- //prepare output\n- out = new ArrayList<>();\n- for( Entry<MatrixIndexes, MatrixBlock> e : rblk.entrySet() )\n- if( outputEmptyBlocks || !e.getValue().isEmptyBlock(false) ) {\n- e.getValue().examSparsity(); //ensure correct format\n- out.add(new IndexedMatrixValue(e.getKey(),e.getValue()));\n- }\n-\n- return out;\n+ //prepare output (sparsity switch, wrapper)\n+ return rblk.entrySet().stream()\n+ .filter( e -> outputEmptyBlocks || !e.getValue().isEmptyBlock(false))\n+ .map(e -> {e.getValue().examSparsity(); return new IndexedMatrixValue(e.getKey(),e.getValue());})\n+ .collect(Collectors.toList());\n}\n/**\n@@ -1553,45 +1547,25 @@ public class LibMatrixReorg\n}\n}\n- @SuppressWarnings(\"unused\")\n- private static HashMap<MatrixIndexes, MatrixBlock> createAllResultBlocks( Collection<MatrixIndexes> rix,\n- long nnz, MatrixCharacteristics mcIn, MatrixCharacteristics mcOut,\n- boolean rowwise, ArrayList<IndexedMatrixValue> reuse )\n- {\n- HashMap<MatrixIndexes, MatrixBlock> ret = new HashMap<MatrixIndexes,MatrixBlock>();\n- long nBlocks = rix.size();\n- int count = 0;\n+ private static Map<MatrixIndexes, MatrixBlock> createAllResultBlocks(Collection<MatrixIndexes> rix, long nnz, MatrixCharacteristics mcOut) {\n+ return rix.stream().collect(Collectors.toMap(ix -> ix, ix -> createResultBlock(ix, nnz, rix.size(), mcOut)));\n+ }\n- for( MatrixIndexes ix : rix )\n- {\n+ private static MatrixBlock createResultBlock(MatrixIndexes ix, long nnz, int nBlocks, MatrixCharacteristics mcOut) {\n//compute indexes\nlong bi = ix.getRowIndex();\nlong bj = ix.getColumnIndex();\nint lbrlen = UtilFunctions.computeBlockSize(mcOut.getRows(), bi, mcOut.getRowsPerBlock());\nint lbclen = UtilFunctions.computeBlockSize(mcOut.getCols(), bj, mcOut.getColsPerBlock());\n-\n+ if( lbrlen<1 || lbclen<1 )\n+ throw new DMLRuntimeException(\"Computed block dimensions (\"+bi+\",\"+bj+\" -> \"+lbrlen+\",\"+lbclen+\") are invalid!\");\n//create result block\nint estnnz = (int) (nnz/nBlocks); //force initial capacity per row to 1, for many blocks\nboolean sparse = MatrixBlock.evalSparseFormatInMemory(lbrlen, lbclen, estnnz);\n- MatrixBlock block = null;\n- if( ALLOW_BLOCK_REUSE && reuse!=null && !reuse.isEmpty()) {\n- block = (MatrixBlock) reuse.get(count++).getValue();\n- block.reset(lbrlen, lbclen, sparse, estnnz);\n- }\n- else\n- block = new MatrixBlock(lbrlen, lbclen, sparse, estnnz);\n-\n- if( lbrlen<1 || lbclen<1 )\n- throw new RuntimeException(\"Computed block dimensions (\"+bi+\",\"+bj+\" -> \"+lbrlen+\",\"+lbclen+\") are invalid!\");\n-\n- ret.put(ix, block);\n- }\n-\n- return ret;\n+ return new MatrixBlock(lbrlen, lbclen, sparse, estnnz);\n}\n- private static void reshapeDense( MatrixBlock in, long row_offset, long col_offset,\n- HashMap<MatrixIndexes,MatrixBlock> rix,\n+ private static void reshapeDense( MatrixBlock in, long row_offset, long col_offset, Map<MatrixIndexes,MatrixBlock> rix,\nMatrixCharacteristics mcIn, MatrixCharacteristics mcOut, boolean rowwise ) {\nif( in.isEmptyBlock(false) )\nreturn;\n@@ -1622,14 +1596,12 @@ public class LibMatrixReorg\n//cleanup for sparse blocks\nif( !rowwise && mcIn.getRows() > 1 ) {\n- for( MatrixBlock block : rix.values() )\n- if( block.sparse )\n- block.sortSparseRows();\n+ rix.values().stream().filter(b -> b.sparse)\n+ .forEach(b -> b.sortSparseRows());\n}\n}\n- private static void reshapeSparse( MatrixBlock in, long row_offset, long col_offset,\n- HashMap<MatrixIndexes,MatrixBlock> rix,\n+ private static void reshapeSparse( MatrixBlock in, long row_offset, long col_offset, Map<MatrixIndexes,MatrixBlock> rix,\nMatrixCharacteristics mcIn, MatrixCharacteristics mcOut, boolean rowwise ) {\nif( in.isEmptyBlock(false) )\nreturn;\n@@ -1639,35 +1611,36 @@ public class LibMatrixReorg\n//append all values to right blocks\nMatrixIndexes ixtmp = new MatrixIndexes();\n- for( int i=0; i<rlen; i++ )\n- {\n- if( !a.isEmpty(i) ) {\n+ for( int i=0; i<rlen; i++ ) {\n+ if( a.isEmpty(i) ) continue;\nlong ai = row_offset+i;\nint apos = a.pos(i);\nint alen = a.size(i);\nint[] aix = a.indexes(i);\ndouble[] avals = a.values(i);\n-\nfor( int j=apos; j<apos+alen; j++ ) {\nlong aj = col_offset+aix[j];\nixtmp = computeResultBlockIndex(ixtmp, ai, aj, mcIn, mcOut, rowwise);\n- MatrixBlock out = rix.get(ixtmp);\n- if( out == null )\n- throw new DMLRuntimeException(\"Missing result block: \"+ixtmp);\n+ MatrixBlock out = getAllocatedBlock(rix, ixtmp);\nixtmp = computeInBlockIndex(ixtmp, ai, aj, mcIn, mcOut, rowwise);\nout.appendValue((int)ixtmp.getRowIndex(),(int)ixtmp.getColumnIndex(), avals[j]);\n}\n}\n- }\n//cleanup for sparse blocks\nif( !rowwise && mcIn.getRows() > 1 ) {\n- for( MatrixBlock block : rix.values() )\n- if( block.sparse )\n- block.sortSparseRows();\n+ rix.values().stream().filter(b -> b.sparse)\n+ .forEach(b -> b.sortSparseRows());\n}\n}\n+ private static MatrixBlock getAllocatedBlock(Map<MatrixIndexes,MatrixBlock> rix, MatrixIndexes ix) {\n+ MatrixBlock out = rix.get(ix);\n+ if( out == null )\n+ throw new DMLRuntimeException(\"Missing result block: \"+ix);\n+ return out;\n+ }\n+\n/**\n* Assumes internal (0-begin) indices ai, aj as input; computes external block indexes (1-begin)\n*\n@@ -1682,25 +1655,27 @@ public class LibMatrixReorg\nprivate static MatrixIndexes computeResultBlockIndex( MatrixIndexes ixout, long ai, long aj,\nMatrixCharacteristics mcIn, MatrixCharacteristics mcOut, boolean rowwise )\n{\n- long tempc = rowwise ? ai*mcIn.getCols()+aj : ai+mcIn.getRows()*aj;\n+ long tempc = computeGlobalCellIndex(mcIn, ai, aj, rowwise);\nlong ci = rowwise ? tempc/mcOut.getCols() : tempc%mcOut.getRows();\nlong cj = rowwise ? tempc%mcOut.getCols() : tempc/mcOut.getRows();\nlong bci = ci/mcOut.getRowsPerBlock() + 1;\nlong bcj = cj/mcOut.getColsPerBlock() + 1;\n- return (ixout != null) ? ixout.setIndexes(bci, bcj) :\n- new MatrixIndexes(bci, bcj);\n+ return ixout.setIndexes(bci, bcj);\n}\nprivate static MatrixIndexes computeInBlockIndex( MatrixIndexes ixout, long ai, long aj,\nMatrixCharacteristics mcIn, MatrixCharacteristics mcOut, boolean rowwise )\n{\n- long tempc = rowwise ? ai*mcIn.getCols()+aj : ai+mcIn.getRows()*aj;\n+ long tempc = computeGlobalCellIndex(mcIn, ai, aj, rowwise);\nlong ci = rowwise ? (tempc/mcOut.getCols())%mcOut.getRowsPerBlock() :\n(tempc%mcOut.getRows())%mcOut.getRowsPerBlock();\nlong cj = rowwise ? (tempc%mcOut.getCols())%mcOut.getColsPerBlock() :\n(tempc/mcOut.getRows())%mcOut.getColsPerBlock();\n- return (ixout != null) ? ixout.setIndexes(ci, cj) :\n- new MatrixIndexes(ci, cj);\n+ return ixout.setIndexes(ci, cj);\n+ }\n+\n+ private static long computeGlobalCellIndex(MatrixCharacteristics mcIn, long ai, long aj, boolean rowwise) {\n+ return rowwise ? ai*mcIn.getCols()+aj : ai+mcIn.getRows()*aj;\n}\nprivate static MatrixBlock removeEmptyRows(MatrixBlock in, MatrixBlock ret, MatrixBlock select, boolean emptyReturn) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -218,7 +218,7 @@ public class OperationsOnMatrixValues\nvalueIn.aggregateUnaryOperations(op, valueOut, brlen, bclen, indexesIn);\n}\n- public static MatrixBlock performAggregateBinary(MatrixIndexes indexes1, MatrixBlock value1, MatrixIndexes indexes2, MatrixBlock value2,\n+ public static MatrixBlock matMult(MatrixIndexes indexes1, MatrixBlock value1, MatrixIndexes indexes2, MatrixBlock value2,\nMatrixIndexes indexesOut, MatrixBlock valueOut, AggregateBinaryOperator op) {\n//compute output index\nindexesOut.setIndexes(indexes1.getRowIndex(), indexes2.getColumnIndex());\n@@ -229,7 +229,7 @@ public class OperationsOnMatrixValues\nreturn value1.aggregateBinaryOperations(indexes1, value1, indexes2, value2, valueOut, op);\n}\n- public static MatrixBlock performAggregateBinaryIgnoreIndexes(MatrixBlock value1, MatrixBlock value2,\n+ public static MatrixBlock matMult(MatrixBlock value1, MatrixBlock value2,\nMatrixBlock valueOut, AggregateBinaryOperator op) {\n//perform on the value\nif( value2 instanceof CompressedMatrixBlock )\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/mapred/MMCJMRReducerWithAggregator.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/mapred/MMCJMRReducerWithAggregator.java", "diff": "@@ -123,15 +123,15 @@ public class MMCJMRReducerWithAggregator extends MMCJMRCombinerReducerBase\n{\n//perform matrix multiplication\nindexesbuffer.setIndexes(tmp.getKey().getRowIndex(), inIndex);\n- OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes((MatrixBlock)tmp.getValue(),\n- (MatrixBlock)inValue, (MatrixBlock)valueBuffer, (AggregateBinaryOperator)aggBinInstruction.getOperator());\n+ OperationsOnMatrixValues.matMult((MatrixBlock)tmp.getValue(), (MatrixBlock)inValue,\n+ (MatrixBlock)valueBuffer, (AggregateBinaryOperator)aggBinInstruction.getOperator());\n}\nelse //right cached\n{\n//perform matrix multiplication\nindexesbuffer.setIndexes(inIndex, tmp.getKey().getColumnIndex());\n- OperationsOnMatrixValues.performAggregateBinaryIgnoreIndexes((MatrixBlock)inValue,\n- (MatrixBlock)tmp.getValue(), (MatrixBlock)valueBuffer, (AggregateBinaryOperator)aggBinInstruction.getOperator());\n+ OperationsOnMatrixValues.matMult((MatrixBlock)inValue, (MatrixBlock)tmp.getValue(),\n+ (MatrixBlock)valueBuffer, (AggregateBinaryOperator)aggBinInstruction.getOperator());\n}\n//aggregate block to output buffer or direct output\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "diff": "@@ -457,7 +457,6 @@ public class LibMatrixReorg\n*\n* @param in indexed matrix value\n* @param mcIn input matrix characteristics\n- * @param out list of indexed matrix values\n* @param mcOut output matrix characteristics\n* @param rowwise if true, reshape by row\n* @param outputEmptyBlocks output blocks with nnz=0\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -3726,7 +3726,8 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\n}\n//compute nnz of output (not maintained due to native calls)\n- dest.recomputeNonZeros();\n+ dest.setNonZeros((getNonZeros() == getLength()) ?\n+ (ru-rl+1) * (cu-cl+1) : dest.recomputeNonZeros());\n}\n@Override\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -42,6 +42,7 @@ import org.apache.sysml.runtime.functionobjects.ValueFunction;\nimport org.apache.sysml.runtime.matrix.operators.ReorgOperator;\nimport org.apache.sysml.runtime.util.CommonThreadPool;\nimport org.apache.sysml.runtime.util.UtilFunctions;\n+import org.apache.sysml.utils.NativeHelper;\n/**\n* MB: Library for matrix multiplications including MM, MV, VV for all\n@@ -557,7 +558,13 @@ public class LibMatrixMult\nret.allocateBlock();\n//core weighted square sum mm computation\n- if( !mW.sparse && !mU.sparse && !mV.sparse && !mU.isEmptyBlock() && !mV.isEmptyBlock() )\n+ boolean allDense = !mW.sparse && !mU.sparse && !mV.sparse\n+ && !mU.isEmptyBlock() && !mV.isEmptyBlock();\n+ if( NativeHelper.isNativeLibraryLoaded() && allDense && (mW.rlen == 1 || mW.clen == 1)\n+ && !LibMatrixNative.isMatMultMemoryBound(mU.rlen, mU.clen, mV.rlen)\n+ && mW.getDenseBlock().isContiguous() && mU.getDenseBlock().isContiguous() && mV.getDenseBlock().isContiguous() )\n+ matrixMultWSigmoidDenseNative(mW, mU, mV, ret, wt);\n+ else if( allDense )\nmatrixMultWSigmoidDense(mW, mU, mV, ret, wt, 0, mW.rlen);\nelse if( mW.sparse && !mU.sparse && !mV.sparse && !mU.isEmptyBlock() && !mV.isEmptyBlock())\nmatrixMultWSigmoidSparseDense(mW, mU, mV, ret, wt, 0, mW.rlen);\n@@ -2384,6 +2391,30 @@ public class LibMatrixMult\ndotProduct(tmp1.getDenseBlockValues(), tmp2.getDenseBlockValues(), mU.clen*mU.clen)));\n}\n+ private static void matrixMultWSigmoidDenseNative(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WSigmoidType wt) {\n+ double[] w = mW.getDenseBlockValues();\n+ double[] c = ret.getDenseBlockValues();\n+ final int m = mW.rlen, n = mW.clen;\n+ final int cd = mU.clen;\n+ boolean flagminus = (wt==WSigmoidType.MINUS || wt==WSigmoidType.LOG_MINUS);\n+ boolean flaglog = (wt==WSigmoidType.LOG || wt==WSigmoidType.LOG_MINUS);\n+\n+ //call native matrix multiplication (only called for single-threaded and matrix-vector\n+ //because this ensures that we can deal with the transpose mV without additional transpose)\n+ if(!NativeHelper.dmmdd(((m==1)?mV:mU).getDenseBlockValues(),\n+ ((m==1)?mU:mV).getDenseBlockValues(), c, (m==1)?n:m, cd, 1, 1) )\n+ throw new DMLRuntimeException(\"Error executing native matrix mult.\");\n+\n+ //compute remaining wsigmoid for all relevant outputs\n+ for(int i=0, ix=0; i<m; i++, ix+=n) {\n+ for(int j=0; j<n; j++) {\n+ double wij = w[ix +j];\n+ //if( wij != 0 )\n+ c[ix+j] = wsigmoid(wij, c[ix+j], flagminus, flaglog);\n+ }\n+ }\n+ }\n+\nprivate static void matrixMultWSigmoidDense(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WSigmoidType wt, int rl, int ru) {\nDenseBlock w = mW.getDenseBlock();\nDenseBlock c = ret.getDenseBlock();\n@@ -3463,6 +3494,16 @@ public class LibMatrixMult\nreturn wij * ((flaglog) ? Math.log(cval) : cval);\n}\n+ private static double wsigmoid(final double wij, final double uvij, final boolean flagminus, final boolean flaglog) {\n+ //compute core sigmoid function\n+ double cval = flagminus ?\n+ 1 / (1 + FastMath.exp(uvij)) :\n+ 1 / (1 + FastMath.exp(-uvij));\n+\n+ //compute weighted output\n+ return wij * ((flaglog) ? Math.log(cval) : cval);\n+ }\n+\nprivate static void wdivmm( final double wij, double[] u, double[] v, double[] c, final int uix, final int vix, final boolean left, final boolean mult, final boolean minus, final int len )\n{\n//compute dot product over ui vj\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixNative.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixNative.java", "diff": "@@ -48,7 +48,7 @@ public class LibMatrixNative\n// We could encapsulate heuristics in this function\n// For now, we only consider matrix-vector operation to be memory bound\n- private static boolean isMatMultMemoryBound(int m1Rlen, int m1Clen, int m2Clen) {\n+ public static boolean isMatMultMemoryBound(int m1Rlen, int m1Clen, int m2Clen) {\nreturn (m1Rlen == 1 || m1Clen == 1 || m2Clen == 1)\n&& (8L*m1Rlen*m1Clen > 16 * LibMatrixMult.L3_CACHESIZE\n|| 8L*m1Clen*m2Clen > 16 * LibMatrixMult.L3_CACHESIZE);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixBincell.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixBincell.java", "diff": "@@ -470,8 +470,10 @@ public class LibMatrixBincell\ndouble bval = b[i];\nif( bval != 0 ) {\nfor( int j=0; j<clen; j++ ) {\n+ double aval = a[aix+j];\n+ if( aval == 0 ) continue;\nindexes[pos] = j;\n- vals[pos] = a[aix+j] * bval;\n+ vals[pos] = aval * bval;\npos++;\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -2584,15 +2584,18 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\nelse\nLibMatrixAgg.cumaggregateUnaryMatrix(this, ret, op);\n}\n- else if(!sparse && !isEmptyBlock(false) && getDenseBlock().isContiguous()\n+ else if(!sparse && !isEmptyBlock(false)\n&& OptimizerUtils.isMaxLocalParallelism(op.getNumThreads())) {\n//note: we apply multi-threading in a best-effort manner here\n//only for expensive operators such as exp, log, sigmoid, because\n//otherwise allocation, read and write anyway dominates\nret.allocateDenseBlock(false);\n- double[] a = getDenseBlockValues();\n- double[] c = ret.getDenseBlockValues();\n- Arrays.parallelSetAll(c, i -> op.fn.execute(a[i]));\n+ DenseBlock a = getDenseBlock();\n+ DenseBlock c = ret.getDenseBlock();\n+ for(int bi=0; bi<a.numBlocks(); bi++) {\n+ double[] avals = a.valuesAt(bi), cvals = c.valuesAt(bi);\n+ Arrays.parallelSetAll(cvals, i -> op.fn.execute(avals[i]));\n+ }\nret.recomputeNonZeros();\n}\nelse {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -2406,6 +2406,9 @@ public class LibMatrixMult\nboolean flagminus = (wt==WSigmoidType.MINUS || wt==WSigmoidType.LOG_MINUS);\nboolean flaglog = (wt==WSigmoidType.LOG || wt==WSigmoidType.LOG_MINUS);\n+ //note: experiments with a fully native implementation of this method (even with #pragma omp simd)\n+ //showed performance regressions compared to this version because we benefit from FastMath.exp\n+\n//call native matrix multiplication (only called for single-threaded and matrix-vector\n//because this ensures that we can deal with the transpose mV without additional transpose)\nif(!NativeHelper.dmmdd(((m==1)?mV:mU).getDenseBlockValues(),\n@@ -2413,12 +2416,13 @@ public class LibMatrixMult\nthrow new DMLRuntimeException(\"Error executing native matrix mult.\");\n//compute remaining wsigmoid for all relevant outputs\n- for(int i=0, ix=0; i<m; i++, ix+=n) {\n- for(int j=0; j<n; j++) {\n- double wij = w[ix +j];\n- //if( wij != 0 )\n- c[ix+j] = wsigmoid(wij, c[ix+j], flagminus, flaglog);\n- }\n+ for( int i=0; i<m*n; i++ ) {\n+ //compute core sigmoid function\n+ double cval = flagminus ?\n+ 1 / (1 + FastMath.exp(c[i])) :\n+ 1 / (1 + FastMath.exp(-c[i]));\n+ //compute weighted output\n+ c[i] = w[i] * ((flaglog) ? Math.log(cval) : cval);\n}\n}\n@@ -3501,16 +3505,6 @@ public class LibMatrixMult\nreturn wij * ((flaglog) ? Math.log(cval) : cval);\n}\n- private static double wsigmoid(final double wij, final double uvij, final boolean flagminus, final boolean flaglog) {\n- //compute core sigmoid function\n- double cval = flagminus ?\n- 1 / (1 + FastMath.exp(uvij)) :\n- 1 / (1 + FastMath.exp(-uvij));\n-\n- //compute weighted output\n- return wij * ((flaglog) ? Math.log(cval) : cval);\n- }\n-\nprivate static void wdivmm( final double wij, double[] u, double[] v, double[] c, final int uix, final int vix, final boolean left, final boolean mult, final boolean minus, final int len )\n{\n//compute dot product over ui vj\n" } ]

[ { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/cholesky.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-source(\"scalable_linalg/triangular_inv.dml\") as inv\n-\n-Choleskey = function(Matrix[double] A, int nb)\n- return(Matrix[double] L) {\n- n = ncol(A)\n-\n- if (n <= nb) {\n- L = cholesky(A)\n- } else {\n- k = as.integer(floor(n/2))\n- A11 = A[1:k,1:k]\n- A21 = A[k+1:n,1:k]\n- A22 = A[k+1:n,k+1:n]\n-\n- L11 = Choleskey(A11, nb)\n- L11inv = inv::U_triangular_inv(t(L11))\n- L21 = A21 %*% L11inv\n- A22 = A22 - L21 %*% t(L21)\n- L22 = Choleskey(A22, nb)\n- L12 = matrix(0, rows=nrow(L11), cols=ncol(L22))\n-\n- L = rbind(cbind(L11, L12), cbind(L21, L22))\n- }\n-}\n" }, { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/inverse.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-source(\"scalable_linalg/triangular_inv.dml\") as tinv\n-source(\"scalable_linalg/qr.dml\") as decomp\n-\n-Inverse = function(Matrix[double] A, int nb)\n- return(Matrix[double] X) {\n- # TODO: Some recent papers discuss Block-Recursive algorithm for\n- # matrix inverse which can be explored instead of QR decomposition\n-\n- [Q, R] = decomp::QR(A, nb)\n-\n- Rinv = tinv::U_triangular_inv(R)\n-\n- X = R %*% t(Q)\n-}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/staging/scalable_linalg/linalg_decomp.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+Cholesky = function(Matrix[double] A, int nb) return(Matrix[double] L) {\n+ n = ncol(A)\n+ if (n <= nb) {\n+ L = cholesky(A)\n+ }\n+ else {\n+ k = as.integer(floor(n/2))\n+ A11 = A[1:k,1:k]\n+ A21 = A[k+1:n,1:k]\n+ A22 = A[k+1:n,k+1:n]\n+\n+ L11 = Cholesky(A11, nb)\n+ L11inv = U_triangular_inv(t(L11))\n+ L21 = A21 %*% L11inv\n+ A22 = A22 - L21 %*% t(L21)\n+ L22 = Cholesky(A22, nb)\n+ L12 = matrix(0, nrow(L11), ncol(L22))\n+\n+ L = rbind(cbind(L11, L12), cbind(L21, L22))\n+ }\n+}\n+\n+Inverse = function(Matrix[double] A, int nb) return(Matrix[double] X) {\n+ # TODO: Some recent papers discuss Block-Recursive algorithm for\n+ # matrix inverse which can be explored instead of QR decomposition\n+\n+ [Q, R] = QR(A, nb)\n+ Rinv = U_triangular_inv(R)\n+ X = R %*% t(Q)\n+}\n+\n+LU = function(Matrix[double] A, int nb) return(Matrix[double] P, Matrix[double] L, Matrix[double] U) {\n+ n = ncol(A)\n+ if (n <= nb) {\n+ [P, L, U] = lu(A)\n+ }\n+ else {\n+ k = as.integer(floor(n/2))\n+ A11 = A[1:k,1:k]\n+ A12 = A[1:k,k+1:n]\n+ A21 = A[k+1:n,1:k]\n+ A22 = A[k+1:n,k+1:n]\n+\n+ [P11, L11, U11] = LU(A11, nb)\n+ L11inv = L_triangular_inv(L11)\n+ U11inv = U_triangular_inv(U11)\n+ U12 = L11inv %*% (P11 %*% A12)\n+ A22 = A22 - A21 %*% U11inv %*% U12\n+ [P22, L22, U22] = LU(A22, nb)\n+ L21 = P22 %*% A21 %*% U11inv\n+\n+ Z12 = matrix(0, nrow(A11), ncol(A22))\n+ Z21 = matrix(0, nrow(A22), ncol(A11))\n+\n+ L = rbind(cbind(L11, Z12), cbind(L21, L22))\n+ U = rbind(cbind(U11, U12), cbind(Z21, U22))\n+ P = rbind(cbind(P11, Z12), cbind(Z21, P22))\n+ }\n+}\n+\n+QR = function(Matrix[double] A, int nb) return(Matrix[double] Q, Matrix[double] R) {\n+ n = ncol(A)\n+ if (n <= nb) {\n+ [H, R] = qr(A)\n+ Q = QfromH(H)\n+ R = R[1:n, 1:n]\n+ }\n+ else {\n+ k = floor(n/2)\n+ A1 = A[,1:k]\n+ A2 = A[,k+1:n]\n+\n+ [Q1, R11] = QR(A1, nb)\n+ R12 = t(Q1) %*% A2\n+ A2 = A2 - Q1 %*% R12\n+ [Q2, R22] = QR(A2, nb)\n+ R21 = matrix(0, nrow(R22), ncol(R11))\n+\n+ Q = cbind(Q1, Q2)\n+ R = rbind(cbind(R11, R12), cbind(R21, R22))\n+ }\n+}\n+\n+# calculate Q from Housholder matrix\n+QfromH = function(Matrix[double] H) return(Matrix[double] Q) {\n+ m = nrow(H);\n+ n = ncol(H);\n+ ones = matrix(1, m, 1);\n+ eye = diag(ones);\n+ Q = eye[,1:n];\n+\n+ for (j in n:1) {\n+ v = H[j:m,j]\n+ b = as.scalar(2/(t(v) %*% v))\n+ Q[j:m, j:n] = Q[j:m, j:n] - (b * v) %*% (t(v) %*% Q[j:m, j:n])\n+ }\n+}\n+\n+Solve = function(Matrix[double] A, Matrix[double] b, int nb) return(Matrix[double] x) {\n+ # TODO: using thin QR may accumulate higher numerical errors.\n+ # Some modifications as suggested by Golub may be explored\n+\n+ [Q, R] = QR(A, nb)\n+ Rinv = U_triangular_inv(R)\n+ x = Rinv %*% t(Q) %*% b\n+}\n+\n+# Inverse of lower triangular matrix\n+L_triangular_inv = function(Matrix[double] L) return(Matrix[double] A) {\n+ n = ncol(L)\n+ if (n == 1) {\n+ A = 1/L[1,1]\n+ }\n+ else if (n == 2) {\n+ A = matrix(0, 2, 2)\n+ A[1,1] = L[2,2]\n+ A[2,2] = L[1,1]\n+ A[2,1] = -L[2,1]\n+ A = A/(as.scalar(L[1,1] * L[2,2]))\n+ }\n+ else {\n+ k = as.integer(floor(n/2))\n+\n+ L11 = L[1:k,1:k]\n+ L21 = L[k+1:n,1:k]\n+ L22 = L[k+1:n,k+1:n]\n+\n+ A11 = L_triangular_inv(L11)\n+ A22 = L_triangular_inv(L22)\n+ A12 = matrix(0, nrow(A11), ncol(A22))\n+ A21 = -A22 %*% L21 %*% A11\n+\n+ A = rbind(cbind(A11, A12), cbind(A21, A22))\n+ }\n+}\n+\n+# Inverse of upper triangular matrix\n+U_triangular_inv = function(Matrix[double] U) return(Matrix[double] A) {\n+ n = ncol(U)\n+ if (n == 1) {\n+ A = 1/U[1,1]\n+ }\n+ else if (n == 2) {\n+ A = matrix(0, 2, 2)\n+ A[1,1] = U[2,2]\n+ A[2,2] = U[1,1]\n+\n+ A[1,2] = -U[1,2]\n+ A = A/(as.scalar(U[1,1] * U[2,2]))\n+ }\n+ else {\n+ k = as.integer(floor(n/2))\n+\n+ U11 = U[1:k,1:k]\n+ U12 = U[1:k,k+1:n]\n+ U22 = U[k+1:n,k+1:n]\n+\n+ A11 = U_triangular_inv(U11)\n+ A22 = U_triangular_inv(U22)\n+ A12 = -A11 %*% U12 %*% A22\n+ A21 = matrix(0, nrow(A22), ncol(A11))\n+\n+ A = rbind(cbind(A11, A12), cbind(A21, A22))\n+ }\n+}\n" }, { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/lu.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-source(\"scalable_linalg/triangular_inv.dml\") as inv\n-\n-LU = function(Matrix[double] A, int nb)\n- return(Matrix[double] P, Matrix[double] L, Matrix[double] U) {\n- n = ncol(A)\n- if (n <= nb) {\n- [P, L, U] = lu(A)\n- } else {\n- k = as.integer(floor(n/2))\n- A11 = A[1:k,1:k]\n- A12 = A[1:k,k+1:n]\n- A21 = A[k+1:n,1:k]\n- A22 = A[k+1:n,k+1:n]\n-\n- [P11, L11, U11] = LU(A11, nb)\n- L11inv = inv::L_triangular_inv(L11)\n- U11inv = inv::U_triangular_inv(U11)\n- U12 = L11inv %*% (P11 %*% A12)\n- A22 = A22 - A21 %*% U11inv %*% U12\n- [P22, L22, U22] = LU(A22, nb)\n- L21 = P22 %*% A21 %*% U11inv\n-\n- Z12 = matrix(0, rows=nrow(A11), cols=ncol(A22))\n- Z21 = matrix(0, rows=nrow(A22), cols=ncol(A11))\n-\n- L = rbind(cbind(L11, Z12), cbind(L21, L22))\n- U = rbind(cbind(U11, U12), cbind(Z21, U22))\n- P = rbind(cbind(P11, Z12), cbind(Z21, P22))\n- }\n-}\n" }, { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/qr.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-# calculate Q from Housholder matrix\n-QfromH = function(Matrix[double] H)\n- return(Matrix[double] Q) {\n- m = nrow(H);\n- n = ncol(H);\n- ones = matrix(1, m, 1);\n- eye = diag(ones);\n- Q = eye[,1:n];\n-\n- for (j in n:1) {\n- v = H[j:m,j]\n- b = as.scalar(2/(t(v) %*% v))\n- Q[j:m, j:n] = Q[j:m, j:n] - (b * v) %*% (t(v) %*% Q[j:m, j:n])\n- }\n-}\n-\n-QR = function(Matrix[double] A, int nb)\n- return(Matrix[double] Q, Matrix[double] R) {\n- n = ncol(A)\n-\n- if (n <= nb) {\n- [H, R] = qr(A)\n- Q = QfromH(H)\n- R = R[1:n, 1:n]\n- }\n- else {\n- k = floor(n/2)\n- A1 = A[,1:k]\n- A2 = A[,k+1:n]\n-\n- [Q1, R11] = QR(A1, nb)\n- R12 = t(Q1) %*% A2\n- A2 = A2 - Q1 %*% R12\n- [Q2, R22] = QR(A2, nb)\n- R21 = matrix(0, rows = nrow(R22), cols = ncol(R11))\n-\n- Q = cbind(Q1, Q2)\n- R = rbind(cbind(R11, R12), cbind(R21, R22))\n- }\n-}\n" }, { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/solve.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-source(\"scalable_linalg/triangular_inv.dml\") as inv\n-source(\"scalable_linalg/qr.dml\") as decomp\n-\n-Solve = function(Matrix[double] A, Matrix[double] b, int nb)\n- return(Matrix[double] x) {\n- # TODO: using thin QR may accumulate higher numerical erros.\n- # Some modifications as suggested by Golub may be explored\n-\n- [Q, R] = decomp::QR(A, nb)\n-\n- Rinv = inv::U_triangular_inv(R)\n-\n- x = Rinv %*% t(Q) %*% b\n-}\n" }, { "change_type": "DELETE", "old_path": "scripts/staging/scalable_linalg/triangular_inv.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-\n-# Inverse of lower triangular matrix\n-L_triangular_inv = function(Matrix[double] L)\n- return(Matrix[double] A) {\n- n = ncol(L)\n-\n- if (n == 1) {\n- A = 1/L[1,1]\n- } else if (n == 2) {\n- A = matrix(0, rows=2, cols=2)\n- A[1,1] = L[2,2]\n- A[2,2] = L[1,1]\n- A[2,1] = -L[2,1]\n- A = A/(as.scalar(L[1,1] * L[2,2]))\n- } else {\n- k = as.integer(floor(n/2))\n-\n- L11 = L[1:k,1:k]\n- L21 = L[k+1:n,1:k]\n- L22 = L[k+1:n,k+1:n]\n-\n- A11 = L_triangular_inv(L11)\n- A22 = L_triangular_inv(L22)\n- A12 = matrix(0, rows=nrow(A11), cols=ncol(A22))\n- A21 = -A22 %*% L21 %*% A11\n-\n- A = rbind(cbind(A11, A12), cbind(A21, A22))\n- }\n- }\n-\n-# Inverse of upper triangular matrix\n-U_triangular_inv = function(Matrix[double] U)\n- return(Matrix[double] A) {\n- n = ncol(U)\n-\n- if (n == 1) {\n- A = 1/U[1,1]\n- } else if (n == 2) {\n- A = matrix(0, rows=2, cols=2)\n- A[1,1] = U[2,2]\n- A[2,2] = U[1,1]\n-\n- A[1,2] = -U[1,2]\n- A = A/(as.scalar(U[1,1] * U[2,2]))\n- } else {\n- k = as.integer(floor(n/2))\n-\n- U11 = U[1:k,1:k]\n- U12 = U[1:k,k+1:n]\n- U22 = U[k+1:n,k+1:n]\n-\n- A11 = U_triangular_inv(U11)\n- A22 = U_triangular_inv(U22)\n- A12 = -A11 %*% U12 %*% A22\n- A21 = matrix(0, rows=nrow(A22), cols=ncol(A11))\n-\n- A = rbind(cbind(A11, A12), cbind(A21, A22))\n- }\n- }\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/BinaryMatrixMatrixCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/BinaryMatrixMatrixCPInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.cp;\n-import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.matrix.data.LibCommonsMath;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\n@@ -35,11 +34,12 @@ public class BinaryMatrixMatrixCPInstruction extends BinaryCPInstruction {\n@Override\npublic void processInstruction(ExecutionContext ec) {\n- String opcode = getOpcode();\n- if ( LibCommonsMath.isSupportedMatrixMatrixOperation(opcode) ) {\n+ if ( LibCommonsMath.isSupportedMatrixMatrixOperation(getOpcode()) ) {\nMatrixBlock solution = LibCommonsMath.matrixMatrixOperations(\n- ec.getMatrixObject(input1.getName()), (MatrixObject)ec.getVariable(input2.getName()), opcode);\n+ ec.getMatrixInput(input1.getName()), ec.getMatrixInput(input2.getName()), getOpcode());\nec.setMatrixOutput(output.getName(), solution, getExtendedOpcode());\n+ ec.releaseMatrixInput(input1.getName());\n+ ec.releaseMatrixInput(input2.getName());\nreturn;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/MultiReturnBuiltinCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/MultiReturnBuiltinCPInstruction.java", "diff": "@@ -24,7 +24,6 @@ import java.util.ArrayList;\nimport org.apache.sysml.parser.Expression.DataType;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n-import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.instructions.InstructionUtils;\nimport org.apache.sysml.runtime.matrix.data.LibCommonsMath;\n@@ -98,16 +97,12 @@ public class MultiReturnBuiltinCPInstruction extends ComputationCPInstruction {\n@Override\npublic void processInstruction(ExecutionContext ec) {\n- String opcode = getOpcode();\n- MatrixObject mo = ec.getMatrixObject(input1.getName());\n- MatrixBlock[] out = null;\n-\n- if(LibCommonsMath.isSupportedMultiReturnOperation(opcode))\n- out = LibCommonsMath.multiReturnOperations(mo, opcode);\n- else\n- throw new DMLRuntimeException(\"Invalid opcode in MultiReturnBuiltin instruction: \" + opcode);\n-\n+ if(!LibCommonsMath.isSupportedMultiReturnOperation(getOpcode()))\n+ throw new DMLRuntimeException(\"Invalid opcode in MultiReturnBuiltin instruction: \" + getOpcode());\n+ MatrixBlock in = ec.getMatrixInput(input1.getName());\n+ MatrixBlock[] out = LibCommonsMath.multiReturnOperations(in, getOpcode());\n+ ec.releaseMatrixInput(input1.getName(), getExtendedOpcode());\nfor(int i=0; i < _outputs.size(); i++) {\nec.setMatrixOutput(_outputs.get(i).getName(), out[i], getExtendedOpcode());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/UnaryMatrixCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/UnaryMatrixCPInstruction.java", "diff": "@@ -32,21 +32,22 @@ public class UnaryMatrixCPInstruction extends UnaryCPInstruction {\n@Override\npublic void processInstruction(ExecutionContext ec) {\n- String output_name = output.getName();\n- String opcode = getOpcode();\n- if(LibCommonsMath.isSupportedUnaryOperation(opcode)) {\n- MatrixBlock retBlock = LibCommonsMath.unaryOperations(ec.getMatrixObject(input1.getName()),getOpcode());\n- ec.setMatrixOutput(output_name, retBlock, getExtendedOpcode());\n+ MatrixBlock inBlock = ec.getMatrixInput(input1.getName(), getExtendedOpcode());\n+ MatrixBlock retBlock = null;\n+\n+ if(LibCommonsMath.isSupportedUnaryOperation(getOpcode())) {\n+ retBlock = LibCommonsMath.unaryOperations(inBlock, getOpcode());\n+ ec.releaseMatrixInput(input1.getName(), getExtendedOpcode());\n}\nelse {\nUnaryOperator u_op = (UnaryOperator) _optr;\n- MatrixBlock inBlock = ec.getMatrixInput(input1.getName(), getExtendedOpcode());\n- MatrixBlock retBlock = (MatrixBlock) (inBlock.unaryOperations(u_op, new MatrixBlock()));\n+ retBlock = (MatrixBlock) (inBlock.unaryOperations(u_op, new MatrixBlock()));\nec.releaseMatrixInput(input1.getName(), getExtendedOpcode());\n// Ensure right dense/sparse output representation (guarded by released input memory)\nif( checkGuardedRepresentationChange(inBlock, retBlock) )\nretBlock.examSparsity();\n- ec.setMatrixOutput(output_name, retBlock, getExtendedOpcode());\n}\n+\n+ ec.setMatrixOutput(output.getName(), retBlock, getExtendedOpcode());\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "diff": "@@ -28,7 +28,6 @@ import org.apache.commons.math3.linear.QRDecomposition;\nimport org.apache.commons.math3.linear.RealMatrix;\nimport org.apache.commons.math3.linear.SingularValueDecomposition;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n-import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.util.DataConverter;\n/**\n@@ -56,7 +55,7 @@ public class LibCommonsMath\nreturn ( opcode.equals(\"solve\") );\n}\n- public static MatrixBlock unaryOperations(MatrixObject inj, String opcode) {\n+ public static MatrixBlock unaryOperations(MatrixBlock inj, String opcode) {\nArray2DRowRealMatrix matrixInput = DataConverter.convertToArray2DRowRealMatrix(inj);\nif(opcode.equals(\"inverse\"))\nreturn computeMatrixInverse(matrixInput);\n@@ -65,7 +64,7 @@ public class LibCommonsMath\nreturn null;\n}\n- public static MatrixBlock[] multiReturnOperations(MatrixObject in, String opcode) {\n+ public static MatrixBlock[] multiReturnOperations(MatrixBlock in, String opcode) {\nif(opcode.equals(\"qr\"))\nreturn computeQR(in);\nelse if (opcode.equals(\"lu\"))\n@@ -77,7 +76,7 @@ public class LibCommonsMath\nreturn null;\n}\n- public static MatrixBlock matrixMatrixOperations(MatrixObject in1, MatrixObject in2, String opcode) {\n+ public static MatrixBlock matrixMatrixOperations(MatrixBlock in1, MatrixBlock in2, String opcode) {\nif(opcode.equals(\"solve\")) {\nif (in1.getNumRows() != in1.getNumColumns())\nthrow new DMLRuntimeException(\"The A matrix, in solve(A,b) should have squared dimensions.\");\n@@ -93,7 +92,7 @@ public class LibCommonsMath\n* @param in2 matrix object 2\n* @return matrix block\n*/\n- private static MatrixBlock computeSolve(MatrixObject in1, MatrixObject in2) {\n+ private static MatrixBlock computeSolve(MatrixBlock in1, MatrixBlock in2) {\nArray2DRowRealMatrix matrixInput = DataConverter.convertToArray2DRowRealMatrix(in1);\nArray2DRowRealMatrix vectorInput = DataConverter.convertToArray2DRowRealMatrix(in2);\n@@ -116,7 +115,7 @@ public class LibCommonsMath\n* @param in matrix object\n* @return array of matrix blocks\n*/\n- private static MatrixBlock[] computeQR(MatrixObject in) {\n+ private static MatrixBlock[] computeQR(MatrixBlock in) {\nArray2DRowRealMatrix matrixInput = DataConverter.convertToArray2DRowRealMatrix(in);\n// Perform QR decomposition\n@@ -137,7 +136,7 @@ public class LibCommonsMath\n* @param in matrix object\n* @return array of matrix blocks\n*/\n- private static MatrixBlock[] computeLU(MatrixObject in) {\n+ private static MatrixBlock[] computeLU(MatrixBlock in) {\nif ( in.getNumRows() != in.getNumColumns() ) {\nthrow new DMLRuntimeException(\"LU Decomposition can only be done on a square matrix. Input matrix is rectangular (rows=\" + in.getNumRows() + \", cols=\"+ in.getNumColumns() +\")\");\n}\n@@ -165,7 +164,7 @@ public class LibCommonsMath\n* @param in matrix object\n* @return array of matrix blocks\n*/\n- private static MatrixBlock[] computeEigen(MatrixObject in) {\n+ private static MatrixBlock[] computeEigen(MatrixBlock in) {\nif ( in.getNumRows() != in.getNumColumns() ) {\nthrow new DMLRuntimeException(\"Eigen Decomposition can only be done on a square matrix. Input matrix is rectangular (rows=\" + in.getNumRows() + \", cols=\"+ in.getNumColumns() +\")\");\n}\n@@ -216,7 +215,7 @@ public class LibCommonsMath\n* @param in Input matrix\n* @return An array containing U, Sigma & V\n*/\n- private static MatrixBlock[] computeSvd(MatrixObject in) {\n+ private static MatrixBlock[] computeSvd(MatrixBlock in) {\nArray2DRowRealMatrix matrixInput = DataConverter.convertToArray2DRowRealMatrix(in);\nSingularValueDecomposition svd = new SingularValueDecomposition(matrixInput);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/util/DataConverter.java", "new_path": "src/main/java/org/apache/sysml/runtime/util/DataConverter.java", "diff": "@@ -31,7 +31,6 @@ import java.util.Map.Entry;\nimport org.apache.commons.math3.linear.Array2DRowRealMatrix;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n-import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.instructions.cp.BooleanObject;\nimport org.apache.sysml.runtime.io.MatrixReader;\nimport org.apache.sysml.runtime.io.MatrixReaderFactory;\n@@ -780,10 +779,8 @@ public class DataConverter\n* @param mo matrix object\n* @return matrix as a commons-math3 Array2DRowRealMatrix\n*/\n- public static Array2DRowRealMatrix convertToArray2DRowRealMatrix(MatrixObject mo) {\n- MatrixBlock mb = mo.acquireRead();\n+ public static Array2DRowRealMatrix convertToArray2DRowRealMatrix(MatrixBlock mb) {\ndouble[][] data = DataConverter.convertToDoubleMatrix(mb);\n- mo.release();\nreturn new Array2DRowRealMatrix(data, false);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/InstructionUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/InstructionUtils.java", "diff": "@@ -87,6 +87,7 @@ import org.apache.sysml.runtime.instructions.gpu.GPUInstruction.GPUINSTRUCTION_T\nimport org.apache.sysml.runtime.instructions.mr.MRInstruction.MRType;\nimport org.apache.sysml.runtime.instructions.spark.SPInstruction.SPType;\nimport org.apache.sysml.runtime.matrix.data.LibCommonsMath;\n+import org.apache.sysml.runtime.matrix.operators.AggregateBinaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateTernaryOperator;\nimport org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator;\n@@ -865,4 +866,9 @@ public class InstructionUtils\n|| WeightedUnaryMM.OPCODE.equalsIgnoreCase(opcode) //mapwumm\n|| WeightedUnaryMMR.OPCODE.equalsIgnoreCase(opcode); //redwumm\n}\n+\n+ public static AggregateBinaryOperator getMatMultOperator(int k) {\n+ AggregateOperator agg = new AggregateOperator(0, Plus.getPlusFnObject());\n+ return new AggregateBinaryOperator(Multiply.getMultiplyFnObject(), agg, k);\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/AutomatedTestBase.java", "new_path": "src/test/java/org/apache/sysml/test/integration/AutomatedTestBase.java", "diff": "@@ -53,6 +53,7 @@ import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.CSVFileFormatProperties;\nimport org.apache.sysml.runtime.matrix.data.FrameBlock;\nimport org.apache.sysml.runtime.matrix.data.InputInfo;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex;\nimport org.apache.sysml.runtime.matrix.data.OutputInfo;\nimport org.apache.sysml.runtime.util.DataConverter;\n@@ -504,6 +505,11 @@ public abstract class AutomatedTestBase\nreturn matrix;\n}\n+ protected double[][] writeInputMatrixWithMTD(String name, MatrixBlock matrix, boolean bIncludeR) {\n+ double[][] data = DataConverter.convertToDoubleMatrix(matrix);\n+ return writeInputMatrixWithMTD(name, data, bIncludeR);\n+ }\n+\nprotected double[][] writeInputMatrixWithMTD(String name, double[][] matrix, boolean bIncludeR) {\nMatrixCharacteristics mc = new MatrixCharacteristics(matrix.length, matrix[0].length, OptimizerUtils.DEFAULT_BLOCKSIZE, OptimizerUtils.DEFAULT_BLOCKSIZE, -1);\nreturn writeInputMatrixWithMTD(name, matrix, bIncludeR, mc);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/applications/dml/ScalableDecompositionTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.test.integration.applications.dml;\n+\n+import java.util.HashMap;\n+\n+import org.junit.Test;\n+import org.apache.sysml.api.DMLScript;\n+import org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\n+import org.apache.sysml.hops.OptimizerUtils;\n+import org.apache.sysml.lops.LopProperties.ExecType;\n+import org.apache.sysml.lops.MMTSJ.MMTSJType;\n+import org.apache.sysml.runtime.instructions.InstructionUtils;\n+import org.apache.sysml.runtime.matrix.data.LibCommonsMath;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+import org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.integration.TestConfiguration;\n+import org.apache.sysml.test.utils.TestUtils;\n+\n+public class ScalableDecompositionTest extends AutomatedTestBase\n+{\n+ private final static String TEST_NAME1 = \"ScalableDecomposition\";\n+ private final static String TEST_DIR = \"applications/decomp/\";\n+ private final static String TEST_CLASS_DIR =\n+ TEST_DIR + ScalableDecompositionTest.class.getSimpleName() + \"/\";\n+\n+ private final static int rows = 1362;\n+ private final static int cols = 1362;\n+ private final static int blen = 200;\n+ private final static double eps = 1e-7;\n+\n+ private enum DecompType {\n+ CHOLESKY, LU, QR, SOLVE, INVERSE\n+ }\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[] { \"C\",\"D\",\"E\" }));\n+ }\n+\n+ @Test\n+ public void testCholeskyCP() {\n+ runKMeansTest(TEST_NAME1, DecompType.CHOLESKY, false, ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testCholeskyRewritesCP() {\n+ runKMeansTest(TEST_NAME1, DecompType.CHOLESKY, true, ExecType.CP);\n+ }\n+\n+// @Test\n+// public void testCholeskySP() {\n+// runKMeansTest(TEST_NAME1, DecompType.CHOLESKY, false, ExecType.SPARK);\n+// }\n+//\n+// @Test\n+// public void testCholeskyRewritesSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.CHOLESKY, true, ExecType.SPARK);\n+// }\n+\n+// @Test\n+// public void testLUDecompCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.LU, false, ExecType.CP);\n+// }\n+//\n+// @Test\n+// public void testLUDecompRewritesCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.LU, true, ExecType.CP);\n+// }\n+\n+// @Test\n+// public void testLUDecompSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.LU, false, ExecType.SPARK);\n+// }\n+//\n+// @Test\n+// public void testLUDecompRewritesSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.LU, true, ExecType.SPARK);\n+// }\n+\n+// @Test\n+// public void testQRDecompCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.QR, false, ExecType.CP);\n+// }\n+//\n+// @Test\n+// public void testQRDecompRewritesCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.QR, true, ExecType.CP);\n+// }\n+\n+// @Test\n+// public void testQRDecompSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.QR, false, ExecType.SPARK);\n+// }\n+//\n+// @Test\n+// public void testQRDecompRewritesSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.QR, true, ExecType.SPARK);\n+// }\n+\n+// @Test\n+// public void testSolveCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, false, ExecType.CP);\n+// }\n+//\n+// @Test\n+// public void testSolveRewritesCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, true, ExecType.CP);\n+// }\n+\n+// @Test\n+// public void testSolveSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, false, ExecType.SPARK);\n+// }\n+//\n+// @Test\n+// public void testSolveRewritesSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, true, ExecType.SPARK);\n+// }\n+\n+// @Test\n+// public void testInverseCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, false, ExecType.CP);\n+// }\n+//\n+// @Test\n+// public void testInverseRewritesCP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, true, ExecType.CP);\n+// }\n+\n+// @Test\n+// public void testInverseSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, false, ExecType.SPARK);\n+// }\n+//\n+// @Test\n+// public void testInverseRewritesSP() {\n+// runKMeansTest(TEST_NAME1, DecompType.SOLVE, true, ExecType.SPARK);\n+// }\n+\n+\n+ private void runKMeansTest(String testname, DecompType type, boolean rewrites, ExecType instType)\n+ {\n+ boolean oldFlag1 = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n+ boolean oldFlag2 = OptimizerUtils.ALLOW_INTER_PROCEDURAL_ANALYSIS;\n+ RUNTIME_PLATFORM platformOld = rtplatform;\n+ switch( instType ){\n+ case SPARK: rtplatform = RUNTIME_PLATFORM.SPARK; break;\n+ default: rtplatform = RUNTIME_PLATFORM.HYBRID_SPARK; break;\n+ }\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK || rtplatform == RUNTIME_PLATFORM.HYBRID_SPARK )\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+\n+ try\n+ {\n+ TestConfiguration config = getTestConfiguration(testname);\n+ loadTestConfiguration(config);\n+\n+ OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION = rewrites;\n+ OptimizerUtils.ALLOW_INTER_PROCEDURAL_ANALYSIS = rewrites;\n+\n+ fullDMLScriptName = SCRIPT_DIR + TEST_DIR + testname + \".dml\";\n+ programArgs = new String[]{\"-explain\",\"-args\",\n+ String.valueOf(type.ordinal()), String.valueOf(blen),\n+ input(\"A\"), input(\"B\"), output(\"C\"), output(\"D\"), output(\"E\") };\n+\n+ switch( type ) {\n+ case CHOLESKY: {\n+ MatrixBlock A = MatrixBlock.randOperations(rows, cols, 1.0, -5, 10, \"uniform\", 7);\n+ MatrixBlock AtA = A.transposeSelfMatrixMultOperations(new MatrixBlock(), MMTSJType.LEFT);\n+ writeInputMatrixWithMTD(\"A\", AtA, false);\n+ runTest(true, false, null, -1);\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(\"C\");\n+ MatrixBlock C2 = LibCommonsMath.unaryOperations(AtA, \"cholesky\");\n+ TestUtils.compareMatrices(dmlfile, C2, eps);\n+ break;\n+ }\n+ case SOLVE: {\n+ MatrixBlock A = MatrixBlock.randOperations(rows, cols, 1.0, -5, 10, \"uniform\", 7);\n+ MatrixBlock b = MatrixBlock.randOperations(cols, 1, 1.0, -1, 1, \"uniform\", 3);\n+ MatrixBlock y = A.aggregateBinaryOperations(A, b, new MatrixBlock(), InstructionUtils.getMatMultOperator(1));\n+ writeInputMatrixWithMTD(\"A\", A, false);\n+ writeInputMatrixWithMTD(\"B\", y, false);\n+ runTest(true, false, null, -1);\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(\"C\");\n+ MatrixBlock C2 = LibCommonsMath.matrixMatrixOperations(A, b, \"solve\");\n+ TestUtils.compareMatrices(dmlfile, C2, eps);\n+ break;\n+ }\n+ case LU: {\n+ MatrixBlock A = MatrixBlock.randOperations(rows, cols, 1.0, -5, 10, \"uniform\", 7);\n+ writeInputMatrixWithMTD(\"A\", A, false);\n+ runTest(true, false, null, -1);\n+ MatrixBlock[] C = LibCommonsMath.multiReturnOperations(A, \"lu\");\n+ String[] outputs = new String[]{\"C\",\"D\",\"E\"};\n+ for(int i=0; i<outputs.length; i++) {\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(outputs[i]);\n+ TestUtils.compareMatrices(dmlfile, C[i], eps);\n+ }\n+ break;\n+ }\n+ case QR: {\n+ MatrixBlock A = MatrixBlock.randOperations(rows, cols, 1.0, -5, 10, \"uniform\", 7);\n+ writeInputMatrixWithMTD(\"A\", A, false);\n+ runTest(true, false, null, -1);\n+ MatrixBlock[] C = LibCommonsMath.multiReturnOperations(A, \"qr\");\n+ String[] outputs = new String[]{\"C\",\"D\",\"E\"};\n+ for(int i=0; i<outputs.length; i++) {\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(outputs[i]);\n+ TestUtils.compareMatrices(dmlfile, C[i], eps);\n+ }\n+ break;\n+ }\n+ case INVERSE: {\n+ MatrixBlock A = MatrixBlock.randOperations(rows, cols, 1.0, -5, 10, \"uniform\", 7);\n+ writeInputMatrixWithMTD(\"A\", A, false);\n+ runTest(true, false, null, -1);\n+ HashMap<CellIndex, Double> dmlfile = readDMLMatrixFromHDFS(\"C\");\n+ MatrixBlock C2 = LibCommonsMath.unaryOperations(A, \"inverse\");\n+ TestUtils.compareMatrices(dmlfile, C2, eps);\n+ break;\n+ }\n+ }\n+ }\n+ finally {\n+ rtplatform = platformOld;\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = sparkConfigOld;\n+ OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION = oldFlag1;\n+ OptimizerUtils.ALLOW_INTER_PROCEDURAL_ANALYSIS = oldFlag2;\n+ }\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/utils/TestUtils.java", "new_path": "src/test/java/org/apache/sysml/test/utils/TestUtils.java", "diff": "@@ -62,6 +62,7 @@ import org.apache.sysml.runtime.matrix.data.MatrixCell;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysml.runtime.matrix.data.OutputInfo;\nimport org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex;\n+import org.apache.sysml.runtime.util.DataConverter;\nimport org.apache.sysml.runtime.util.UtilFunctions;\nimport org.apache.sysml.test.integration.AutomatedTestBase;\n@@ -693,22 +694,18 @@ public class TestUtils\nassertEquals(expected, actual);\n}\n- /**\n- *\n- * @param m1\n- * @param m2\n- * @param tolerance\n- * @param name1\n- * @param name2\n- * @param ignoreNaN\n- * @return\n- */\npublic static boolean compareMatrices(HashMap<CellIndex, Double> m1, HashMap<CellIndex, Double> m2,\ndouble tolerance, String name1, String name2)\n{\nreturn compareMatrices(m1, m2, tolerance, name1, name2, false);\n}\n+ public static void compareMatrices(HashMap<CellIndex, Double> m1, MatrixBlock m2, double tolerance) {\n+ double[][] ret1 = convertHashMapToDoubleArray(m1);\n+ double[][] ret2 = DataConverter.convertToDoubleMatrix(m2);\n+ compareMatrices(ret1, ret2, m2.getNumRows(), m2.getNumColumns(), tolerance);\n+ }\n+\n/**\n* Compares two matrices given as HashMaps. The matrix containing more nnz\n* is iterated and each cell value compared against the corresponding cell\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/applications/decomp/ScalableDecomposition.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+source(\"scripts/staging/scalable_linalg/linalg_decomp.dml\") as decomp\n+\n+A = read($3);\n+\n+if( $1 == 0 ) {\n+ C = decomp::Cholesky(A, $2);\n+}\n+else if( $1 == 1 ) {\n+ [C, D, E] = decomp::LU(A, $2);\n+}\n+else if( $1 == 2 ) {\n+ [C, D] = decomp::QR(A, $2);\n+}\n+else if( $1 == 3 ) {\n+ B = read($4);\n+ C = decomp::Solve(A, B, $2);\n+}\n+else if( $1 == 4 ) {\n+ C = decomp::Inverse(A, $2);\n+}\n+\n+write(C, $5);\n+if( exists(D) )\n+ write(D, $6)\n+if( exists(E) )\n+ write(E, $7)\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/applications/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/applications/ZPackageSuite.java", "diff": "@@ -27,7 +27,6 @@ import org.junit.runners.Suite;\n* they should not be run in parallel. */\n@RunWith(Suite.class)\[email protected]({\n-\n// .applications.dml package\norg.apache.sysml.test.integration.applications.dml.ApplyTransformDMLTest.class,\norg.apache.sysml.test.integration.applications.dml.ArimaDMLTest.class,\n@@ -45,6 +44,7 @@ import org.junit.runners.Suite;\norg.apache.sysml.test.integration.applications.dml.NaiveBayesDMLTest.class,\norg.apache.sysml.test.integration.applications.dml.NaiveBayesParforDMLTest.class,\norg.apache.sysml.test.integration.applications.dml.PageRankDMLTest.class,\n+ org.apache.sysml.test.integration.applications.dml.ScalableDecompositionTest.class,\norg.apache.sysml.test.integration.applications.dml.WelchTDMLTest.class,\n// .applications.pydml package\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/frame/FrameConverterTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/frame/FrameConverterTest.java", "diff": "@@ -324,7 +324,7 @@ public class FrameConverterTest extends AutomatedTestBase\n{\ntry\n{\n- MatrixCharacteristics mcMatrix = new MatrixCharacteristics(rows, schema.length, 1000, 1000, 0);\n+ MatrixCharacteristics mcMatrix = new MatrixCharacteristics(rows, schema.length, 1000, 1000, -1);\nMatrixCharacteristics mcFrame = new MatrixCharacteristics(rows, schema.length, -1, -1, -1);\nMatrixBlock matrixBlock1 = null;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/AggregateUnarySPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/AggregateUnarySPInstruction.java", "diff": "@@ -34,6 +34,7 @@ import org.apache.sysml.runtime.instructions.InstructionUtils;\nimport org.apache.sysml.runtime.instructions.cp.CPOperand;\nimport org.apache.sysml.runtime.instructions.spark.functions.AggregateDropCorrectionFunction;\nimport org.apache.sysml.runtime.instructions.spark.functions.FilterDiagBlocksFunction;\n+import org.apache.sysml.runtime.instructions.spark.functions.FilterNonEmptyBlocksFunction;\nimport org.apache.sysml.runtime.instructions.spark.utils.RDDAggregateUtils;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\n@@ -91,6 +92,9 @@ public class AggregateUnarySPInstruction extends UnarySPInstruction {\n//perform aggregation if necessary and put output into symbol table\nif( _aggtype == SparkAggType.SINGLE_BLOCK )\n{\n+ if( auop.sparseSafe )\n+ out = out.filter(new FilterNonEmptyBlocksFunction());\n+\nJavaRDD<MatrixBlock> out2 = out.map(\nnew RDDUAggFunction2(auop, mc.getRowsPerBlock(), mc.getColsPerBlock()));\nMatrixBlock out3 = RDDAggregateUtils.aggStable(out2, aggop);\n@@ -164,7 +168,7 @@ public class AggregateUnarySPInstruction extends UnarySPInstruction {\n/**\n* Similar to RDDUAggFunction but single output block.\n*/\n- private static class RDDUAggFunction2 implements Function<Tuple2<MatrixIndexes, MatrixBlock>, MatrixBlock>\n+ public static class RDDUAggFunction2 implements Function<Tuple2<MatrixIndexes, MatrixBlock>, MatrixBlock>\n{\nprivate static final long serialVersionUID = 2672082409287856038L;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/RDDAggregateUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/utils/RDDAggregateUtils.java", "diff": "@@ -26,13 +26,16 @@ import org.apache.spark.api.java.function.Function2;\nimport org.apache.sysml.lops.PartialAggregate.CorrectionLocationType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.functionobjects.KahanPlus;\n+import org.apache.sysml.runtime.instructions.InstructionUtils;\nimport org.apache.sysml.runtime.instructions.cp.KahanObject;\n+import org.apache.sysml.runtime.instructions.spark.AggregateUnarySPInstruction.RDDUAggFunction2;\nimport org.apache.sysml.runtime.instructions.spark.data.CorrMatrixBlock;\nimport org.apache.sysml.runtime.instructions.spark.data.RowMatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysml.runtime.matrix.data.OperationsOnMatrixValues;\nimport org.apache.sysml.runtime.matrix.operators.AggregateOperator;\n+import org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator;\n/**\n* Collection of utility methods for aggregating binary block rdds. As a general\n@@ -83,8 +86,8 @@ public class RDDAggregateUtils\n//stable sum of blocks per key, by passing correction blocks along with aggregates\nJavaPairRDD<MatrixIndexes, CorrMatrixBlock> tmp =\nin.combineByKey( new CreateCorrBlockCombinerFunction(deepCopyCombiner),\n- new MergeSumBlockValueFunction(),\n- new MergeSumBlockCombinerFunction(), numPartitions );\n+ new MergeSumBlockValueFunction(deepCopyCombiner),\n+ new MergeSumBlockCombinerFunction(deepCopyCombiner), numPartitions );\n//strip-off correction blocks from\nJavaPairRDD<MatrixIndexes, MatrixBlock> out =\n@@ -94,13 +97,18 @@ public class RDDAggregateUtils\nreturn out;\n}\n- public static JavaPairRDD<MatrixIndexes, Double> sumCellsByKeyStable( JavaPairRDD<MatrixIndexes, Double> in )\n+\n+ public static JavaPairRDD<MatrixIndexes, Double> sumCellsByKeyStable( JavaPairRDD<MatrixIndexes, Double> in ) {\n+ return sumCellsByKeyStable(in, in.getNumPartitions());\n+ }\n+\n+ public static JavaPairRDD<MatrixIndexes, Double> sumCellsByKeyStable( JavaPairRDD<MatrixIndexes, Double> in, int numParts )\n{\n//stable sum of blocks per key, by passing correction blocks along with aggregates\nJavaPairRDD<MatrixIndexes, KahanObject> tmp =\nin.combineByKey( new CreateCellCombinerFunction(),\nnew MergeSumCellValueFunction(),\n- new MergeSumCellCombinerFunction() );\n+ new MergeSumCellCombinerFunction(), numParts);\n//strip-off correction blocks from\nJavaPairRDD<MatrixIndexes, Double> out =\n@@ -166,6 +174,12 @@ public class RDDAggregateUtils\nreturn out;\n}\n+ public static double max(JavaPairRDD<MatrixIndexes, MatrixBlock> in) {\n+ AggregateUnaryOperator auop = InstructionUtils.parseBasicAggregateUnaryOperator(\"uamax\");\n+ MatrixBlock tmp = aggStable(in.map(new RDDUAggFunction2(auop, -1, -1)), auop.aggOp);\n+ return tmp.quickGetValue(0, 0);\n+ }\n+\n/**\n* Merges disjoint data of all blocks per key.\n*\n@@ -258,6 +272,12 @@ public class RDDAggregateUtils\nprivate AggregateOperator _op = new AggregateOperator(0, KahanPlus.getKahanPlusFnObject(), true, CorrectionLocationType.NONE);\n+ private final boolean _deep;\n+\n+ public MergeSumBlockValueFunction(boolean deep) {\n+ _deep = deep;\n+ }\n+\n@Override\npublic CorrMatrixBlock call(CorrMatrixBlock arg0, MatrixBlock arg1)\nthrows Exception\n@@ -270,12 +290,12 @@ public class RDDAggregateUtils\nMatrixBlock corr = arg0.getCorrection();\n//correction block allocation on demand\n- if( corr == null )\n+ if( corr == null && !arg1.isEmptyBlock(false) )\ncorr = new MatrixBlock(value.getNumRows(), value.getNumColumns(), false);\n//aggregate other input and maintain corrections\n//(existing value and corr are used in place)\n- OperationsOnMatrixValues.incrementalAggregation(value, corr, arg1, _op, false);\n+ OperationsOnMatrixValues.incrementalAggregation(value, corr, arg1, _op, false, _deep);\nreturn arg0.set(value, corr);\n}\n}\n@@ -285,6 +305,11 @@ public class RDDAggregateUtils\nprivate static final long serialVersionUID = 7664941774566119853L;\nprivate AggregateOperator _op = new AggregateOperator(0, KahanPlus.getKahanPlusFnObject(), true, CorrectionLocationType.NONE);\n+ private final boolean _deep;\n+\n+ public MergeSumBlockCombinerFunction(boolean deep) {\n+ _deep = deep;\n+ }\n@Override\npublic CorrMatrixBlock call(CorrMatrixBlock arg0, CorrMatrixBlock arg1)\n@@ -298,12 +323,13 @@ public class RDDAggregateUtils\n//correction block allocation on demand (but use second if exists)\nif( corr == null ) {\ncorr = (arg1.getCorrection()!=null) ? arg1.getCorrection() :\n+ value2.isEmptyBlock(false) || (!_deep && value1.isEmptyBlock(false)) ? null :\nnew MatrixBlock(value1.getNumRows(), value1.getNumColumns(), false);\n}\n//aggregate other input and maintain corrections\n//(existing value and corr are used in place)\n- OperationsOnMatrixValues.incrementalAggregation(value1, corr, value2, _op, false);\n+ OperationsOnMatrixValues.incrementalAggregation(value1, corr, value2, _op, false, _deep);\nreturn arg0.set(value1, corr);\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/io/ReaderBinaryBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/io/ReaderBinaryBlock.java", "diff": "@@ -113,7 +113,7 @@ public class ReaderBinaryBlock extends MatrixReader\n//where ultra-sparse deserialization only reuses CSR blocks\nMatrixBlock value = new MatrixBlock(brlen, bclen, sparse);\nif( sparse ) {\n- value.allocateAndResetSparseRowsBlock(true, SparseBlock.Type.CSR);\n+ value.allocateAndResetSparseBlock(true, SparseBlock.Type.CSR);\nvalue.getSparseBlock().allocate(0, brlen*bclen);\n}\nreturn value;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/CM_N_COVCell.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/CM_N_COVCell.java", "diff": "@@ -98,7 +98,7 @@ public class CM_N_COVCell extends MatrixValue implements WritableComparable\n@Override\npublic void incrementalAggregate(AggregateOperator aggOp,\n- MatrixValue correction, MatrixValue newWithCorrection) {\n+ MatrixValue correction, MatrixValue newWithCorrection, boolean deep) {\nthrow new RuntimeException(\"operation not supported for CM_N_COVCell\");\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixAgg.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixAgg.java", "diff": "@@ -122,12 +122,23 @@ public class LibMatrixAgg\n* @param in input matrix\n* @param aggVal current aggregate values (in/out)\n* @param aggCorr current aggregate correction (in/out)\n+ * @param deep deep copy flag\n*/\n- public static void aggregateBinaryMatrix(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr) {\n+ public static void aggregateBinaryMatrix(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr, boolean deep) {\n//Timing time = new Timing(true);\n//boolean saggVal = aggVal.sparse, saggCorr = aggCorr.sparse;\n//long naggVal = aggVal.nonZeros, naggCorr = aggCorr.nonZeros;\n+ //common empty block handling\n+ if( in.isEmptyBlock(false) ) {\n+ return;\n+ }\n+ if( !deep && aggVal.isEmptyBlock(false) ) {\n+ //shallow copy without correction allocation\n+ aggVal.copyShallow(in);\n+ return;\n+ }\n+\n//ensure MCSR instead of CSR for update in-place\nif( aggVal.sparse && aggVal.isAllocated() && aggVal.getSparseBlock() instanceof SparseBlockCSR )\naggVal.sparseBlock = SparseBlockFactory.copySparseBlock(SparseBlock.Type.MCSR, aggVal.getSparseBlock(), true);\n@@ -977,9 +988,6 @@ public class LibMatrixAgg\n}\nprivate static void aggregateBinaryMatrixAllDense(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr) {\n- if( in.denseBlock==null || in.isEmptyBlock(false) )\n- return;\n-\n//allocate output arrays (if required)\naggVal.allocateDenseBlock(); //should always stay in dense\naggCorr.allocateDenseBlock(); //should always stay in dense\n@@ -1011,9 +1019,6 @@ public class LibMatrixAgg\n}\nprivate static void aggregateBinaryMatrixSparseDense(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr) {\n- if( in.isEmptyBlock(false) )\n- return;\n-\n//allocate output arrays (if required)\naggVal.allocateDenseBlock(); //should always stay in dense\naggCorr.allocateDenseBlock(); //should always stay in dense\n@@ -1055,9 +1060,6 @@ public class LibMatrixAgg\n}\nprivate static void aggregateBinaryMatrixSparseGeneric(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr) {\n- if( in.isEmptyBlock(false) )\n- return;\n-\nSparseBlock a = in.getSparseBlock();\nKahanObject buffer1 = new KahanObject(0, 0);\n@@ -1095,9 +1097,6 @@ public class LibMatrixAgg\n}\nprivate static void aggregateBinaryMatrixDenseGeneric(MatrixBlock in, MatrixBlock aggVal, MatrixBlock aggCorr) {\n- if( in.denseBlock==null || in.isEmptyBlock(false) )\n- return;\n-\nfinal int m = in.rlen;\nfinal int n = in.clen;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixCell.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixCell.java", "diff": "@@ -266,7 +266,7 @@ public class MatrixCell extends MatrixValue implements WritableComparable, Seria\n@Override\npublic void incrementalAggregate(AggregateOperator aggOp,\n- MatrixValue correction, MatrixValue newWithCorrection) {\n+ MatrixValue correction, MatrixValue newWithCorrection, boolean deep) {\nthrow new DMLRuntimeException(\"MatrixCell.incrementalAggregate should never be called\");\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixValue.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixValue.java", "diff": "@@ -127,8 +127,11 @@ public abstract class MatrixValue implements WritableComparable\npublic abstract MatrixValue unaryOperations(UnaryOperator op, MatrixValue result);\n- public abstract void incrementalAggregate(AggregateOperator aggOp, MatrixValue correction,\n- MatrixValue newWithCorrection);\n+ public void incrementalAggregate(AggregateOperator aggOp, MatrixValue correction, MatrixValue newWithCorrection) {\n+ incrementalAggregate(aggOp, correction, newWithCorrection, true);\n+ }\n+\n+ public abstract void incrementalAggregate(AggregateOperator aggOp, MatrixValue correction, MatrixValue newWithCorrection, boolean deep);\npublic abstract void incrementalAggregate(AggregateOperator aggOp, MatrixValue newWithCorrection);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/OperationsOnMatrixValues.java", "diff": "@@ -196,12 +196,18 @@ public class OperationsOnMatrixValues\n}\npublic static void incrementalAggregation(MatrixValue valueAgg, MatrixValue correction, MatrixValue valueAdd,\n- AggregateOperator op, boolean imbededCorrection)\n+ AggregateOperator op, boolean imbededCorrection) {\n+ incrementalAggregation(valueAgg, correction, valueAdd, op, imbededCorrection, true);\n+ }\n+\n+\n+ public static void incrementalAggregation(MatrixValue valueAgg, MatrixValue correction, MatrixValue valueAdd,\n+ AggregateOperator op, boolean imbededCorrection, boolean deep)\n{\nif(op.correctionExists)\n{\nif(!imbededCorrection || op.correctionLocation==CorrectionLocationType.NONE)\n- valueAgg.incrementalAggregate(op, correction, valueAdd);\n+ valueAgg.incrementalAggregate(op, correction, valueAdd, deep);\nelse\nvalueAgg.incrementalAggregate(op, valueAdd);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "diff": "@@ -308,6 +308,7 @@ public class ParForProgramBlock extends ForProgramBlock\npublic static final boolean FORCE_CP_ON_REMOTE_MR = true; // compile body to CP if exec type forced to MR\npublic static final boolean LIVEVAR_AWARE_EXPORT = true; // export only read variables according to live variable analysis\npublic static final boolean RESET_RECOMPILATION_FLAGs = true;\n+ public static final boolean ALLOW_BROADCAST_INPUTS = false; // enables to broadcast inputs for remote_spark\npublic static final String PARFOR_FNAME_PREFIX = \"/parfor/\";\npublic static final String PARFOR_MR_TASKS_TMP_FNAME = PARFOR_FNAME_PREFIX + \"%ID%_MR_taskfile\";\n@@ -1042,8 +1043,8 @@ public class ParForProgramBlock extends ForProgramBlock\nexportMatricesToHDFS(ec);\n// Step 3) submit Spark parfor job (no lazy evaluation, since collect on result)\n- //MatrixObject colocatedDPMatrixObj = (_colocatedDPMatrix!=null)? (MatrixObject)ec.getVariable(_colocatedDPMatrix) : null;\n- RemoteParForJobReturn ret = RemoteParForSpark.runJob(_ID, program, clsMap, tasks, ec, _enableCPCaching, _numThreads);\n+ RemoteParForJobReturn ret = RemoteParForSpark.runJob(_ID, program,\n+ clsMap, tasks, ec, _resultVars, _enableCPCaching, _numThreads);\nif( _monitor )\nStatisticMonitor.putPFStat(_ID, Stat.PARFOR_WAIT_EXEC_T, time.stop());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "diff": "@@ -196,6 +196,11 @@ public class ExecutionContext {\nreturn (MatrixObject) dat;\n}\n+ public boolean isFrameObject(String varname) {\n+ Data dat = getVariable(varname);\n+ return (dat!= null && dat instanceof FrameObject);\n+ }\n+\npublic FrameObject getFrameObject(CPOperand input) {\nreturn getFrameObject(input.getName());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/SparkExecutionContext.java", "diff": "@@ -506,9 +506,47 @@ public class SparkExecutionContext extends ExecutionContext\nreturn rdd;\n}\n+ public Broadcast<CacheBlock> broadcastVariable(CacheableData<CacheBlock> cd) {\n+ long t0 = DMLScript.STATISTICS ? System.nanoTime() : 0;\n+ Broadcast<CacheBlock> brBlock = null;\n+\n+ // reuse existing non partitioned broadcast handle\n+ if (cd.getBroadcastHandle() != null && cd.getBroadcastHandle().isNonPartitionedBroadcastValid()) {\n+ brBlock = cd.getBroadcastHandle().getNonPartitionedBroadcast();\n+ }\n+\n+ if (brBlock == null) {\n+ //create new broadcast handle (never created, evicted)\n+ // account for overwritten invalid broadcast (e.g., evicted)\n+ if (cd.getBroadcastHandle() != null)\n+ CacheableData.addBroadcastSize(-cd.getBroadcastHandle().getNonPartitionedBroadcastSize());\n+\n+ // read the matrix block\n+ CacheBlock cb = cd.acquireRead();\n+ cd.release();\n+\n+ // broadcast a non-empty frame whose size is smaller than 2G\n+ if (cb.getExactSerializedSize() > 0 && cb.getExactSerializedSize() <= Integer.MAX_VALUE) {\n+ brBlock = getSparkContext().broadcast(cb);\n+ // create the broadcast handle if the matrix or frame has never been broadcasted\n+ if (cd.getBroadcastHandle() == null) {\n+ cd.setBroadcastHandle(new BroadcastObject<>());\n+ }\n+ cd.getBroadcastHandle().setNonPartitionedBroadcast(brBlock,\n+ OptimizerUtils.estimateSize(cd.getMatrixCharacteristics()));\n+ CacheableData.addBroadcastSize(cd.getBroadcastHandle().getNonPartitionedBroadcastSize());\n+\n+ if (DMLScript.STATISTICS) {\n+ Statistics.accSparkBroadCastTime(System.nanoTime() - t0);\n+ Statistics.incSparkBroadcastCount(1);\n+ }\n+ }\n+ }\n+ return brBlock;\n+ }\n+\n@SuppressWarnings(\"unchecked\")\n- public PartitionedBroadcast<MatrixBlock> getBroadcastForVariable( String varname )\n- {\n+ public PartitionedBroadcast<MatrixBlock> getBroadcastForVariable(String varname) {\n//NOTE: The memory consumption of this method is the in-memory size of the\n//matrix object plus the partitioned size in 1k-1k blocks. Since the call\n//to broadcast happens after the matrix object has been released, the memory\n@@ -524,18 +562,15 @@ public class SparkExecutionContext extends ExecutionContext\nPartitionedBroadcast<MatrixBlock> bret = null;\n//reuse existing broadcast handle\n- if( mo.getBroadcastHandle()!=null\n- && mo.getBroadcastHandle().isValid() )\n- {\n- bret = mo.getBroadcastHandle().getBroadcast();\n+ if (mo.getBroadcastHandle() != null && mo.getBroadcastHandle().isPartitionedBroadcastValid()) {\n+ bret = mo.getBroadcastHandle().getPartitionedBroadcast();\n}\n//create new broadcast handle (never created, evicted)\n- if( bret == null )\n- {\n+ if (bret == null) {\n//account for overwritten invalid broadcast (e.g., evicted)\nif (mo.getBroadcastHandle() != null)\n- CacheableData.addBroadcastSize(-mo.getBroadcastHandle().getSize());\n+ CacheableData.addBroadcastSize(-mo.getBroadcastHandle().getPartitionedBroadcastSize());\n//obtain meta data for matrix\nint brlen = (int) mo.getNumRowsPerBlock();\n@@ -554,18 +589,20 @@ public class SparkExecutionContext extends ExecutionContext\n//create coarse-grained partitioned broadcasts\nif (numParts > 1) {\nArrays.parallelSetAll(ret, i -> createPartitionedBroadcast(pmb, numPerPart, i));\n- }\n- else { //single partition\n+ } else { //single partition\nret[0] = getSparkContext().broadcast(pmb);\nif (!isLocalMaster())\npmb.clearBlocks();\n}\nbret = new PartitionedBroadcast<>(ret, mo.getMatrixCharacteristics());\n- BroadcastObject<MatrixBlock> bchandle = new BroadcastObject<>(bret,\n+ // create the broadcast handle if the matrix or frame has never been broadcasted\n+ if (mo.getBroadcastHandle() == null) {\n+ mo.setBroadcastHandle(new BroadcastObject<MatrixBlock>());\n+ }\n+ mo.getBroadcastHandle().setPartitionedBroadcast(bret,\nOptimizerUtils.estimatePartitionedSizeExactSparsity(mo.getMatrixCharacteristics()));\n- mo.setBroadcastHandle(bchandle);\n- CacheableData.addBroadcastSize(bchandle.getSize());\n+ CacheableData.addBroadcastSize(mo.getBroadcastHandle().getPartitionedBroadcastSize());\n}\nif (DMLScript.STATISTICS) {\n@@ -577,8 +614,7 @@ public class SparkExecutionContext extends ExecutionContext\n}\n@SuppressWarnings(\"unchecked\")\n- public PartitionedBroadcast<FrameBlock> getBroadcastForFrameVariable( String varname)\n- {\n+ public PartitionedBroadcast<FrameBlock> getBroadcastForFrameVariable(String varname) {\nlong t0 = DMLScript.STATISTICS ? System.nanoTime() : 0;\nFrameObject fo = getFrameObject(varname);\n@@ -586,18 +622,15 @@ public class SparkExecutionContext extends ExecutionContext\nPartitionedBroadcast<FrameBlock> bret = null;\n//reuse existing broadcast handle\n- if( fo.getBroadcastHandle()!=null\n- && fo.getBroadcastHandle().isValid() )\n- {\n- bret = fo.getBroadcastHandle().getBroadcast();\n+ if (fo.getBroadcastHandle() != null && fo.getBroadcastHandle().isPartitionedBroadcastValid()) {\n+ bret = fo.getBroadcastHandle().getPartitionedBroadcast();\n}\n//create new broadcast handle (never created, evicted)\n- if( bret == null )\n- {\n+ if (bret == null) {\n//account for overwritten invalid broadcast (e.g., evicted)\nif (fo.getBroadcastHandle() != null)\n- CacheableData.addBroadcastSize(-fo.getBroadcastHandle().getSize());\n+ CacheableData.addBroadcastSize(-fo.getBroadcastHandle().getPartitionedBroadcastSize());\n//obtain meta data for frame\nint bclen = (int) fo.getNumColumns();\n@@ -616,18 +649,19 @@ public class SparkExecutionContext extends ExecutionContext\n//create coarse-grained partitioned broadcasts\nif (numParts > 1) {\nArrays.parallelSetAll(ret, i -> createPartitionedBroadcast(pmb, numPerPart, i));\n- }\n- else { //single partition\n+ } else { //single partition\nret[0] = getSparkContext().broadcast(pmb);\nif (!isLocalMaster())\npmb.clearBlocks();\n}\nbret = new PartitionedBroadcast<>(ret, fo.getMatrixCharacteristics());\n- BroadcastObject<FrameBlock> bchandle = new BroadcastObject<>(bret,\n+ if (fo.getBroadcastHandle() == null) {\n+ fo.setBroadcastHandle(new BroadcastObject<FrameBlock>());\n+ }\n+ fo.getBroadcastHandle().setPartitionedBroadcast(bret,\nOptimizerUtils.estimatePartitionedSizeExactSparsity(fo.getMatrixCharacteristics()));\n- fo.setBroadcastHandle(bchandle);\n- CacheableData.addBroadcastSize(bchandle.getSize());\n+ CacheableData.addBroadcastSize(fo.getBroadcastHandle().getPartitionedBroadcastSize());\n}\nif (DMLScript.STATISTICS) {\n@@ -1124,11 +1158,20 @@ public class SparkExecutionContext extends ExecutionContext\n_parRDDs.deregisterRDD(rddID);\n}\nelse if( lob instanceof BroadcastObject ) {\n- PartitionedBroadcast pbm = ((BroadcastObject)lob).getBroadcast();\n- if( pbm != null ) //robustness for evictions\n- for( Broadcast<PartitionedBlock> bc : pbm.getBroadcasts() )\n+ BroadcastObject bob = (BroadcastObject) lob;\n+ // clean the partitioned broadcast\n+ if (bob.isPartitionedBroadcastValid()) {\n+ PartitionedBroadcast pbm = bob.getPartitionedBroadcast();\n+ if( pbm != null ) //robustness evictions\n+ pbm.destroy();\n+ }\n+ // clean the non-partitioned broadcast\n+ if (((BroadcastObject) lob).isNonPartitionedBroadcastValid()) {\n+ Broadcast<CacheableData> bc = bob.getNonPartitionedBroadcast();\n+ if( bc != null ) //robustness evictions\ncleanupBroadcastVariable(bc);\n- CacheableData.addBroadcastSize(-((BroadcastObject)lob).getSize());\n+ }\n+ CacheableData.addBroadcastSize(-bob.getNonPartitionedBroadcastSize());\n}\n//recursively process lineage children\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/CachedReuseVariables.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/CachedReuseVariables.java", "diff": "@@ -23,8 +23,15 @@ import java.lang.ref.SoftReference;\nimport java.util.Collection;\nimport java.util.HashMap;\nimport java.util.HashSet;\n+import java.util.Map;\n+import java.util.Map.Entry;\n+import org.apache.spark.broadcast.Broadcast;\nimport org.apache.sysml.runtime.controlprogram.LocalVariableMap;\n+import org.apache.sysml.runtime.controlprogram.ParForProgramBlock;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheableData;\n+import org.apache.sysml.runtime.instructions.cp.Data;\npublic class CachedReuseVariables\n{\n@@ -34,10 +41,21 @@ public class CachedReuseVariables\n_data = new HashMap<>();\n}\n- public synchronized void reuseVariables(long pfid, LocalVariableMap vars, Collection<String> blacklist) {\n+ public synchronized boolean containsVars(long pfid) {\n+ return _data.containsKey(pfid);\n+ }\n+\n+ @SuppressWarnings(\"unused\")\n+ public synchronized void reuseVariables(long pfid, LocalVariableMap vars, Collection<String> blacklist, Map<String, Broadcast<CacheBlock>> _brInputs) {\n+\n+ //fetch the broadcast variables\n+ if (ParForProgramBlock.ALLOW_BROADCAST_INPUTS && !containsVars(pfid)) {\n+ loadBroadcastVariables(vars, _brInputs);\n+ }\n+\n//check for existing reuse map\nLocalVariableMap tmp = null;\n- if( _data.containsKey(pfid) )\n+ if( containsVars(pfid) )\ntmp = _data.get(pfid).get();\n//build reuse map if not created yet or evicted\n@@ -57,4 +75,15 @@ public class CachedReuseVariables\npublic synchronized void clearVariables(long pfid) {\n_data.remove(pfid);\n}\n+\n+ @SuppressWarnings(\"unchecked\")\n+ private static void loadBroadcastVariables(LocalVariableMap variables, Map<String, Broadcast<CacheBlock>> brInputs) {\n+ for( Entry<String, Broadcast<CacheBlock>> e : brInputs.entrySet() ) {\n+ Data d = variables.get(e.getKey());\n+ CacheableData<CacheBlock> cdcb = (CacheableData<CacheBlock>) d;\n+ cdcb.acquireModify(e.getValue().getValue());\n+ cdcb.setEmptyStatus(); // avoid eviction\n+ cdcb.refreshMetaData();\n+ }\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParForSpark.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParForSpark.java", "diff": "package org.apache.sysml.runtime.controlprogram.parfor;\n+import java.util.ArrayList;\nimport java.util.HashMap;\nimport java.util.List;\n+import java.util.Map;\n+import java.util.Set;\n+import java.util.stream.Collectors;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\nimport org.apache.spark.api.java.JavaSparkContext;\n+import org.apache.spark.broadcast.Broadcast;\nimport org.apache.spark.util.LongAccumulator;\n+import org.apache.sysml.parser.ParForStatementBlock;\n+import org.apache.sysml.parser.ParForStatementBlock.ResultVar;\n+import org.apache.sysml.runtime.controlprogram.ParForProgramBlock;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheableData;\n+import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\n+import org.apache.sysml.runtime.instructions.cp.Data;\n+import org.apache.sysml.runtime.instructions.cp.ScalarObject;\nimport scala.Tuple2;\nimport org.apache.sysml.api.DMLScript;\n@@ -47,7 +60,6 @@ import org.apache.sysml.utils.Statistics;\n* pre-aggregation by overwriting partial task results with pre-paggregated results from subsequent\n* iterations)\n*\n- * TODO broadcast variables if possible\n* TODO reducebykey on variable names\n*/\npublic class RemoteParForSpark\n@@ -58,7 +70,7 @@ public class RemoteParForSpark\nprivate static final IDSequence _jobID = new IDSequence();\npublic static RemoteParForJobReturn runJob(long pfid, String prog, HashMap<String, byte[]> clsMap,\n- List<Task> tasks, ExecutionContext ec, boolean cpCaching, int numMappers)\n+ List<Task> tasks, ExecutionContext ec, ArrayList<ResultVar> resultVars, boolean cpCaching, int numMappers)\n{\nString jobname = \"ParFor-ESP\";\nlong t0 = DMLScript.STATISTICS ? System.nanoTime() : 0;\n@@ -75,10 +87,16 @@ public class RemoteParForSpark\nif( InfrastructureAnalyzer.isLocalMode() )\nRemoteParForSparkWorker.cleanupCachedVariables(jobid);\n+ // broadcast the inputs except the result variables\n+ Map<String, Broadcast<CacheBlock>> brInputs = null;\n+ if (ParForProgramBlock.ALLOW_BROADCAST_INPUTS) {\n+ brInputs = broadcastInputs(sec, resultVars);\n+ }\n+\n//run remote_spark parfor job\n//(w/o lazy evaluation to fit existing parfor framework, e.g., result merge)\nList<Tuple2<Long,String>> out = sc.parallelize(tasks, tasks.size()) //create rdd of parfor tasks\n- .flatMapToPair(new RemoteParForSparkWorker(jobid, prog, clsMap, cpCaching, aTasks, aIters))\n+ .flatMapToPair(new RemoteParForSparkWorker(jobid, prog, clsMap, cpCaching, aTasks, aIters, brInputs))\n.collect(); //execute and get output handles\n//de-serialize results\n@@ -97,4 +115,25 @@ public class RemoteParForSpark\nreturn ret;\n}\n+\n+ @SuppressWarnings(\"unchecked\")\n+ private static Map<String, Broadcast<CacheBlock>> broadcastInputs(SparkExecutionContext sec, ArrayList<ParForStatementBlock.ResultVar> resultVars) {\n+ LocalVariableMap inputs = sec.getVariables();\n+ // exclude the result variables\n+ // TODO use optimizer-picked list of amenable objects (e.g., size constraints)\n+ Set<String> retVars = resultVars.stream()\n+ .map(v -> v._name).collect(Collectors.toSet());\n+ Set<String> brVars = inputs.keySet().stream()\n+ .filter(v -> !retVars.contains(v)).collect(Collectors.toSet());\n+\n+ // construct broadcast objects\n+ Map<String, Broadcast<CacheBlock>> result = new HashMap<>();\n+ for (String key : brVars) {\n+ Data var = sec.getVariable(key);\n+ if ((var instanceof ScalarObject) || (var instanceof MatrixObject && ((MatrixObject) var).isPartitioned()))\n+ continue;\n+ result.put(key, sec.broadcastVariable((CacheableData<CacheBlock>) var));\n+ }\n+ return result;\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParForSparkWorker.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParForSparkWorker.java", "diff": "@@ -23,13 +23,16 @@ import java.io.IOException;\nimport java.util.Collection;\nimport java.util.HashMap;\nimport java.util.Iterator;\n+import java.util.Map;\nimport java.util.Map.Entry;\nimport java.util.stream.Collectors;\nimport org.apache.spark.TaskContext;\nimport org.apache.spark.api.java.function.PairFlatMapFunction;\n+import org.apache.spark.broadcast.Broadcast;\nimport org.apache.spark.util.LongAccumulator;\nimport org.apache.sysml.runtime.codegen.CodegenUtils;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.parfor.stat.InfrastructureAnalyzer;\nimport org.apache.sysml.runtime.controlprogram.parfor.util.IDHandler;\n@@ -53,7 +56,9 @@ public class RemoteParForSparkWorker extends ParWorker implements PairFlatMapFun\nprivate final LongAccumulator _aTasks;\nprivate final LongAccumulator _aIters;\n- public RemoteParForSparkWorker(long jobid, String program, HashMap<String, byte[]> clsMap, boolean cpCaching, LongAccumulator atasks, LongAccumulator aiters) {\n+ private final Map<String, Broadcast<CacheBlock>> _brInputs;\n+\n+ public RemoteParForSparkWorker(long jobid, String program, HashMap<String, byte[]> clsMap, boolean cpCaching, LongAccumulator atasks, LongAccumulator aiters, Map<String, Broadcast<CacheBlock>> brInputs) {\n_jobid = jobid;\n_prog = program;\n_clsMap = clsMap;\n@@ -62,6 +67,7 @@ public class RemoteParForSparkWorker extends ParWorker implements PairFlatMapFun\n//setup spark accumulators\n_aTasks = atasks;\n_aIters = aiters;\n+ _brInputs = brInputs;\n}\n@Override\n@@ -108,7 +114,7 @@ public class RemoteParForSparkWorker extends ParWorker implements PairFlatMapFun\n_ec.pinVariables(_ec.getVarList()); //avoid cleanup of shared inputs\nCollection<String> blacklist = UtilFunctions.asSet(_resultVars.stream()\n.map(v -> v._name).collect(Collectors.toList()), _ec.getVarListPartitioned());\n- reuseVars.reuseVariables(_jobid, _ec.getVariables(), blacklist);\n+ reuseVars.reuseVariables(_jobid, _ec.getVariables(), blacklist, _brInputs);\n//init and register-cleanup of buffer pool (in parfor spark, multiple tasks might\n//share the process-local, i.e., per executor, buffer pool; hence we synchronize\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/BroadcastObject.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/data/BroadcastObject.java", "diff": "@@ -24,31 +24,60 @@ import java.lang.ref.SoftReference;\nimport org.apache.spark.broadcast.Broadcast;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\n-public class BroadcastObject<T extends CacheBlock> extends LineageObject\n-{\n+public class BroadcastObject<T extends CacheBlock> extends LineageObject {\n//soft reference storage for graceful cleanup in case of memory pressure\n- protected final SoftReference<PartitionedBroadcast<T>> _bcHandle;\n- private final long _size;\n+ private SoftReference<PartitionedBroadcast<T>> _pbcRef; // partitioned broadcast object reference\n+ private SoftReference<Broadcast<T>> _npbcRef; // non partitioned broadcast object reference\n- public BroadcastObject( PartitionedBroadcast<T> bvar, long size ) {\n+ private long _pbcSize; // partitioned broadcast size\n+ private long _npbcSize; // non-partitioned broadcast size\n+\n+ public BroadcastObject() {\nsuper();\n- _bcHandle = new SoftReference<>(bvar);\n- _size = size;\n+ }\n+\n+ public void setNonPartitionedBroadcast(Broadcast<T> bvar, long size) {\n+ _npbcRef = new SoftReference<>(bvar);\n+ _npbcSize = size;\n+ }\n+\n+ public void setPartitionedBroadcast(PartitionedBroadcast<T> bvar, long size) {\n+ _pbcRef = new SoftReference<>(bvar);\n+ _pbcSize = size;\n}\n@SuppressWarnings(\"rawtypes\")\n- public PartitionedBroadcast getBroadcast() {\n- return _bcHandle.get();\n+ public PartitionedBroadcast getPartitionedBroadcast() {\n+ return _pbcRef.get();\n+ }\n+\n+ public Broadcast<T> getNonPartitionedBroadcast() {\n+ return _npbcRef.get();\n+ }\n+\n+ public long getPartitionedBroadcastSize() {\n+ return _pbcSize;\n+ }\n+\n+ public long getNonPartitionedBroadcastSize() {\n+ return _npbcSize;\n+ }\n+\n+ public boolean isPartitionedBroadcastValid() {\n+ return _pbcRef != null && checkPartitionedBroadcastValid();\n+ }\n+\n+ public boolean isNonPartitionedBroadcastValid() {\n+ return _npbcRef != null && checkNonPartitionedBroadcastValid();\n}\n- public long getSize() {\n- return _size;\n+ private boolean checkNonPartitionedBroadcastValid() {\n+ return _npbcRef.get() != null;\n}\n- public boolean isValid()\n- {\n+ private boolean checkPartitionedBroadcastValid() {\n//check for evicted soft reference\n- PartitionedBroadcast<T> pbm = _bcHandle.get();\n+ PartitionedBroadcast<T> pbm = _pbcRef.get();\nif (pbm == null)\nreturn false;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "diff": "@@ -984,7 +984,7 @@ public class OptimizerUtils\nreturn Math.min(1, nnz1/m) * Math.min(1, nnz2/n);\n}\nelse\n- return (1 - Math.pow(1-sp1*sp2, k) );\n+ return 1 - Math.pow(1-sp1*sp2, k);\n}\npublic static double getLeftIndexingSparsity( long rlen1, long clen1, long nnz1, long rlen2, long clen2, long nnz2 )\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/hops/estim/EstimatorBasicAvg.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.hops.estim;\n+\n+import org.apache.sysml.hops.OptimizerUtils;\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+\n+/**\n+ * Basic average case estimator for matrix sparsity:\n+ * sp = 1 - Math.pow(1-sp1*sp2, k)\n+ */\n+public class EstimatorBasicAvg extends SparsityEstimator\n+{\n+ @Override\n+ public double estim(MMNode root) {\n+ double sp1 = !root.getLeft().isLeaf() ? estim(root.getLeft()) :\n+ OptimizerUtils.getSparsity(root.getLeft().getMatrixCharacteristics());\n+ double sp2 = !root.getRight().isLeaf() ? estim(root.getRight()) :\n+ OptimizerUtils.getSparsity(root.getRight().getMatrixCharacteristics());\n+ return estimIntern(sp1, sp2, root.getRows(), root.getLeft().getCols(), root.getCols());\n+ }\n+\n+ @Override\n+ public double estim(MatrixBlock m1, MatrixBlock m2) {\n+ return estim(m1.getMatrixCharacteristics(), m2.getMatrixCharacteristics());\n+ }\n+\n+ @Override\n+ public double estim(MatrixCharacteristics mc1, MatrixCharacteristics mc2) {\n+ return estimIntern(\n+ OptimizerUtils.getSparsity(mc1), OptimizerUtils.getSparsity(mc2),\n+ mc1.getRows(), mc1.getCols(), mc2.getCols());\n+ }\n+\n+ private double estimIntern(double sp1, double sp2, long m, long k, long n) {\n+ return OptimizerUtils.getMatMultSparsity(sp1, sp2, m, k, n, false);\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/hops/estim/EstimatorBasicWorst.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.hops.estim;\n+\n+import org.apache.sysml.hops.OptimizerUtils;\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+\n+/**\n+ * Basic average case estimator for matrix sparsity:\n+ * sp = Math.min(1, sp1 * k) * Math.min(1, sp2 * k).\n+ *\n+ * Note: for outer-products (i.e., k=1) this worst-case\n+ * estimate is equivalent to the average case estimate and\n+ * the exact output sparsity.\n+ */\n+public class EstimatorBasicWorst extends SparsityEstimator\n+{\n+ @Override\n+ public double estim(MMNode root) {\n+ double sp1 = !root.getLeft().isLeaf() ? estim(root.getLeft()) :\n+ OptimizerUtils.getSparsity(root.getLeft().getMatrixCharacteristics());\n+ double sp2 = !root.getRight().isLeaf() ? estim(root.getRight()) :\n+ OptimizerUtils.getSparsity(root.getRight().getMatrixCharacteristics());\n+ return estimIntern(sp1, sp2, root.getRows(), root.getLeft().getCols(), root.getCols());\n+ }\n+\n+ @Override\n+ public double estim(MatrixBlock m1, MatrixBlock m2) {\n+ return estim(m1.getMatrixCharacteristics(), m2.getMatrixCharacteristics());\n+ }\n+\n+ @Override\n+ public double estim(MatrixCharacteristics mc1, MatrixCharacteristics mc2) {\n+ return estimIntern(\n+ OptimizerUtils.getSparsity(mc1), OptimizerUtils.getSparsity(mc2),\n+ mc1.getRows(), mc1.getCols(), mc2.getCols());\n+ }\n+\n+ private double estimIntern(double sp1, double sp2, long m, long k, long n) {\n+ return OptimizerUtils.getMatMultSparsity(sp1, sp2, m, k, n, true);\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/hops/estim/MMNode.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.hops.estim;\n+\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+\n+/**\n+ * Helper class to represent matrix multiply operators in a DAG\n+ * along with references to its abstract data handles.\n+ */\n+public class MMNode\n+{\n+ private final MMNode _m1;\n+ private final MMNode _m2;\n+ private final MatrixBlock _data;\n+ private final MatrixCharacteristics _mc;\n+\n+ public MMNode(MMNode left, MMNode right) {\n+ _m1 = left;\n+ _m2 = right;\n+ _data = null;\n+ _mc = isLeaf() ? _data.getMatrixCharacteristics() :\n+ new MatrixCharacteristics(_data.getNumRows(),\n+ _data.getNumColumns(), -1, -1);\n+ }\n+\n+ public long getRows() {\n+ return _mc.getRows();\n+ }\n+\n+ public long getCols() {\n+ return _mc.getCols();\n+ }\n+\n+ public MatrixCharacteristics getMatrixCharacteristics() {\n+ return _mc;\n+ }\n+\n+ public MMNode getLeft() {\n+ return _m1;\n+ }\n+\n+ public MMNode getRight() {\n+ return _m2;\n+ }\n+\n+ public boolean isLeaf() {\n+ return _data != null;\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/hops/estim/SparsityEstimator.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.hops.estim;\n+\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+\n+public abstract class SparsityEstimator\n+{\n+ /**\n+ * Estimates the output sparsity of a DAG of matrix multiplications\n+ * for the given operator graph of a single root node.\n+ *\n+ * @param root\n+ * @return\n+ */\n+ public abstract double estim(MMNode root);\n+\n+ /**\n+ * Estimates the output sparsity of a single matrix multiplication\n+ * for the two given matrices.\n+ *\n+ * @param m1 left-hand-side operand\n+ * @param m2 right-hand-side operand\n+ * @return sparsity\n+ */\n+ public abstract double estim(MatrixBlock m1, MatrixBlock m2);\n+\n+ /**\n+ * Estimates the output sparsity of a single matrix multiplication\n+ * for the two given matrices represented by meta data only.\n+ *\n+ * @param mc1 left-hand-side operand\n+ * @param mc2 right-hand-side operand\n+ * @return sparsity\n+ */\n+ public abstract double estim(MatrixCharacteristics mc1, MatrixCharacteristics mc2);\n+\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -463,6 +463,10 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\nreturn OptimizerUtils.getSparsity(rlen, clen, nonZeros);\n}\n+ public MatrixCharacteristics getMatrixCharacteristics() {\n+ return new MatrixCharacteristics(rlen, clen, -1, -1, nonZeros);\n+ }\n+\npublic boolean isVector() {\nreturn (rlen == 1 || clen == 1);\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/estim/OuterProductTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.test.integration.functions.estim;\n+\n+import org.junit.Test;\n+import org.apache.sysml.hops.estim.EstimatorBasicAvg;\n+import org.apache.sysml.hops.estim.EstimatorBasicWorst;\n+import org.apache.sysml.hops.estim.SparsityEstimator;\n+import org.apache.sysml.runtime.instructions.InstructionUtils;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.utils.TestUtils;\n+\n+/**\n+ * This is a basic sanity check for all estimator, which need\n+ * to compute the exact sparsity for the special case of outer products.\n+ */\n+public class OuterProductTest extends AutomatedTestBase\n+{\n+ private final static int m = 1154;\n+ private final static int k = 1;\n+ private final static int n = 900;\n+ private final static double[] case1 = new double[]{0.1, 0.7};\n+ private final static double[] case2 = new double[]{0.6, 0.7};\n+\n+ @Override\n+ public void setUp() {\n+ //do nothing\n+ }\n+\n+ @Test\n+ public void testBasicAvgCase1() {\n+ runSparsityEstimateTest(new EstimatorBasicAvg(), m, n, k, case1);\n+ }\n+\n+ @Test\n+ public void testBasicAvgCase2() {\n+ runSparsityEstimateTest(new EstimatorBasicAvg(), m, n, k, case2);\n+ }\n+\n+ @Test\n+ public void testBasicWorstCase1() {\n+ runSparsityEstimateTest(new EstimatorBasicWorst(), m, n, k, case1);\n+ }\n+\n+ @Test\n+ public void testBasicWorstCase2() {\n+ runSparsityEstimateTest(new EstimatorBasicWorst(), m, n, k, case2);\n+ }\n+\n+ private void runSparsityEstimateTest(SparsityEstimator estim, int m, int k, int n, double[] sp) {\n+ MatrixBlock m1 = MatrixBlock.randOperations(m, k, sp[0], 1, 1, \"uniform\", 3);\n+ MatrixBlock m2 = MatrixBlock.randOperations(k, n, sp[1], 1, 1, \"uniform\", 3);\n+ MatrixBlock m3 = m1.aggregateBinaryOperations(m1, m2,\n+ new MatrixBlock(), InstructionUtils.getMatMultOperator(1));\n+\n+ //compare estimated and real sparsity\n+ double est = estim.estim(m1, m2);\n+ TestUtils.compareScalars(est, m3.getSparsity(), 1e-16);\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/estim/ZPackageSuite.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysml.test.integration.functions.estim;\n+\n+import org.junit.runner.RunWith;\n+import org.junit.runners.Suite;\n+\n+/** Group together the tests in this package into a single suite so that the Maven build\n+ * won't run two of them at once. */\n+@RunWith(Suite.class)\[email protected]({\n+ OuterProductTest.class,\n+})\n+\n+\n+/** This class is just a holder for the above JUnit annotations. */\n+public class ZPackageSuite {\n+\n+}\n" } ]