Repository: systemml Updated Branches: refs/heads/master 6133be23e -> 648eb21d6
http://git-wip-us.apache.org/repos/asf/systemml/blob/648eb21d/src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java ---------------------------------------------------------------------- diff --git a/src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java b/src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java index 16051dc..835e491 100644 --- a/src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java +++ b/src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java @@ -61,6 +61,7 @@ import org.apache.sysml.runtime.functionobjects.RevIndex; import org.apache.sysml.runtime.functionobjects.SortIndex; import org.apache.sysml.runtime.functionobjects.SwapIndex; import org.apache.sysml.runtime.instructions.cp.CM_COV_Object; +import org.apache.sysml.runtime.instructions.cp.CPInstruction; import org.apache.sysml.runtime.instructions.cp.KahanObject; import org.apache.sysml.runtime.instructions.cp.ScalarObject; import org.apache.sysml.runtime.io.IOUtilFunctions; @@ -86,6 +87,7 @@ import org.apache.sysml.runtime.util.FastBufferedDataInputStream; import org.apache.sysml.runtime.util.FastBufferedDataOutputStream; import org.apache.sysml.runtime.util.IndexRange; import org.apache.sysml.runtime.util.UtilFunctions; +import org.apache.sysml.utils.GPUStatistics; import org.apache.sysml.utils.NativeHelper; import org.apache.sysml.utils.Statistics; @@ -992,6 +994,10 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab return evalSparseFormatOnDisk(lrlen, lclen, nonZeros); } + public void examSparsity() throws DMLRuntimeException { + examSparsity(null); + } + /** * Evaluates if this matrix block should be in sparse format in * memory. Depending on the current representation, the state of the @@ -999,10 +1005,11 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab * Note that this consumes for the time of execution memory for both * representations. * + * @param opcode extended opcode * @throws DMLRuntimeException if DMLRuntimeException occurs */ @SuppressWarnings("unused") - public void examSparsity() + public void examSparsity(String opcode) throws DMLRuntimeException { long start = DISPLAY_STATISTICS && DMLScript.STATISTICS ? System.nanoTime() : 0; @@ -1016,9 +1023,9 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab //change representation if required (also done for //empty blocks in order to set representation flags) if( sparse && !sparseDst) - sparseToDense(); + sparseToDense(opcode); else if( !sparse && sparseDst ) - denseToSparse(); + denseToSparse(opcode); Statistics.examSparsityTime += DISPLAY_STATISTICS && DMLScript.STATISTICS ? (System.nanoTime() - start) : 0; } @@ -1075,7 +1082,14 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab // basic block handling functions void denseToSparse() + { + denseToSparse(null); + } + + void denseToSparse(String opcode) { + long t1 = opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS ? System.nanoTime() : 0; + //set target representation sparse = true; @@ -1115,11 +1129,20 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab //update nnz and cleanup dense block nonZeros = nnz; denseBlock = null; + if(opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS) { + long t2 = System.nanoTime(); + GPUStatistics.maintainCPMiscTimes(opcode, CPInstruction.MISC_TIMER_DENSE_TO_SPARSE, t2-t1); + } + } + + public void sparseToDense() throws DMLRuntimeException { + sparseToDense(null); } - public void sparseToDense() + public void sparseToDense(String opcode) throws DMLRuntimeException - { + { + long t1 = opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS ? System.nanoTime() : 0; //set target representation sparse = false; @@ -1153,6 +1176,10 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab //cleanup sparse rows sparseBlock = null; + if(opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS) { + long t2 = System.nanoTime(); + GPUStatistics.maintainCPMiscTimes(opcode, CPInstruction.MISC_TIMER_SPARSE_TO_DENSE, t2-t1); + } } /** @@ -3613,13 +3640,22 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab LibMatrixMult.matrixMultPermute(this, m2, ret1, ret2); } + + public final MatrixBlock leftIndexingOperations(MatrixBlock rhsMatrix, IndexRange ixrange, MatrixBlock ret, UpdateType update) throws DMLRuntimeException { + return leftIndexingOperations(rhsMatrix, ixrange, ret, update, null); + } - public final MatrixBlock leftIndexingOperations(MatrixBlock rhsMatrix, IndexRange ixrange, MatrixBlock ret, UpdateType update) + public final MatrixBlock leftIndexingOperations(MatrixBlock rhsMatrix, IndexRange ixrange, MatrixBlock ret, UpdateType update, String opcode) throws DMLRuntimeException { return leftIndexingOperations( rhsMatrix, (int)ixrange.rowStart, (int)ixrange.rowEnd, - (int)ixrange.colStart, (int)ixrange.colEnd, ret, update); + (int)ixrange.colStart, (int)ixrange.colEnd, ret, update, opcode); + } + + public MatrixBlock leftIndexingOperations(MatrixBlock rhsMatrix, int rl, int ru, + int cl, int cu, MatrixBlock ret, UpdateType update) throws DMLRuntimeException { + return leftIndexingOperations(rhsMatrix, rl, ru, cl, cu, ret, update, null); } /** @@ -3641,7 +3677,7 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab * @throws DMLRuntimeException if DMLRuntimeException occurs */ public MatrixBlock leftIndexingOperations(MatrixBlock rhsMatrix, int rl, int ru, - int cl, int cu, MatrixBlock ret, UpdateType update) + int cl, int cu, MatrixBlock ret, UpdateType update, String opcode) throws DMLRuntimeException { // Check the validity of bounds @@ -3678,9 +3714,9 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab //ensure that the current block adheres to the sparsity estimate //and thus implicitly the memory budget used by the compiler if( result.sparse && !sp ) - result.sparseToDense(); + result.sparseToDense(opcode); else if( !result.sparse && sp ) - result.denseToSparse(); + result.denseToSparse(opcode); //ensure right sparse block representation to prevent serialization if( result.sparse && update != UpdateType.INPLACE_PINNED ) { @@ -3702,6 +3738,8 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab result.quickSetValue(rl, cl, src.quickGetValue(0, 0)); } else { //general case + long t1 = opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS ? System.nanoTime() : 0; + boolean isCSRCopy = false; //handle csr sparse blocks separately to avoid repeated shifting on column-wise access if( !result.isEmptyBlock(false) && result.sparse && result.sparseBlock instanceof SparseBlockCSR ) { SparseBlockCSR sblock = (SparseBlockCSR) result.sparseBlock; @@ -3710,11 +3748,19 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab else //dense sblock.setIndexRange(rl, ru+1, cl, cu+1, src.getDenseBlock(), 0, src.getNumRows()*src.getNumColumns()); result.nonZeros = sblock.size(); + isCSRCopy = true; } //copy submatrix into result else { result.copy(rl, ru, cl, cu, src, true); } + if(opcode != null && DMLScript.STATISTICS && DMLScript.FINEGRAINED_STATISTICS) { + long t2 = System.nanoTime(); + if(isCSRCopy) + GPUStatistics.maintainCPMiscTimes(opcode, CPInstruction.MISC_TIMER_CSR_LIX_COPY, t2-t1); + else + GPUStatistics.maintainCPMiscTimes(opcode, CPInstruction.MISC_TIMER_LIX_COPY, t2-t1); + } } return result; http://git-wip-us.apache.org/repos/asf/systemml/blob/648eb21d/src/main/java/org/apache/sysml/utils/GPUStatistics.java ---------------------------------------------------------------------- diff --git a/src/main/java/org/apache/sysml/utils/GPUStatistics.java b/src/main/java/org/apache/sysml/utils/GPUStatistics.java index 6b85d92..c0f1d5e 100644 --- a/src/main/java/org/apache/sysml/utils/GPUStatistics.java +++ b/src/main/java/org/apache/sysml/utils/GPUStatistics.java @@ -28,6 +28,8 @@ import java.util.List; import java.util.Map; import java.util.concurrent.atomic.AtomicLong; +import org.apache.sysml.api.DMLScript; + /** * Measures performance numbers when GPU mode is enabled * Printed as part of {@link Statistics}. @@ -119,7 +121,7 @@ public class GPUStatistics { */ public synchronized static void maintainCPMiscTimes( String instructionName, String miscTimer, long timeNanos, long incrementCount) { - if (!DISPLAY_STATISTICS) + if (!(DISPLAY_STATISTICS || DMLScript.FINEGRAINED_STATISTICS)) return; HashMap<String, Long> miscTimesMap = _cpInstMiscTime.get(instructionName); http://git-wip-us.apache.org/repos/asf/systemml/blob/648eb21d/src/main/java/org/apache/sysml/utils/Statistics.java ---------------------------------------------------------------------- diff --git a/src/main/java/org/apache/sysml/utils/Statistics.java b/src/main/java/org/apache/sysml/utils/Statistics.java index 808267d..a72b89e 100644 --- a/src/main/java/org/apache/sysml/utils/Statistics.java +++ b/src/main/java/org/apache/sysml/utils/Statistics.java @@ -533,7 +533,7 @@ public class Statistics final String instCol = "Instruction"; final String timeSCol = "Time(s)"; final String countCol = "Count"; - final String gpuCol = "GPU"; + final String gpuCol = "Misc Timers"; StringBuilder sb = new StringBuilder(); int numHittersToDisplay = Math.min(num, len); int maxNumLen = String.valueOf(numHittersToDisplay).length(); @@ -557,7 +557,7 @@ public class Statistics sb.append(String.format( " %" + maxNumLen + "s %-" + maxInstLen + "s %" + maxTimeSLen + "s %" + maxCountLen + "s", numCol, instCol, timeSCol, countCol)); - if (GPUStatistics.DISPLAY_STATISTICS) { + if (GPUStatistics.DISPLAY_STATISTICS || DMLScript.FINEGRAINED_STATISTICS) { sb.append(" "); sb.append(gpuCol); } @@ -575,7 +575,7 @@ public class Statistics (i + 1), instruction, timeSString, count)); // Add the miscellaneous timer info - if (GPUStatistics.DISPLAY_STATISTICS) { + if (GPUStatistics.DISPLAY_STATISTICS || DMLScript.FINEGRAINED_STATISTICS) { sb.append(" "); sb.append(GPUStatistics.getStringForCPMiscTimesPerInstruction(instruction)); }
