ebfdcc9a044d15c0bd9c4355b2c92ff3d6f95024 kent Sun Dec 19 08:50:12 2021 -0800 Parallelizing memMatrixToTsv and moving it to a library. Parallelizing column normalizing. diff --git src/utils/matrixNormalize/matrixNormalize.c src/utils/matrixNormalize/matrixNormalize.c index 5cf4322..df1a08a 100644 --- src/utils/matrixNormalize/matrixNormalize.c +++ src/utils/matrixNormalize/matrixNormalize.c @@ -1,51 +1,46 @@ /* matrixNormalize - Normalize a matrix somehow - make it's columns or rows all sum to one or have vector length one.. */ #include "common.h" #include "linefile.h" #include "hash.h" #include "options.h" #include "vMatrix.h" #include "pthreadDoList.h" -int clThreads = 10; - void usage() /* Explain usage and exit. */ { errAbort( "matrixNormalize - Normalize a matrix somehow - make it's columns or rows all sum to one or have vector length one.\n" "usage:\n" " matrixNormalize direction how inMatrix outMatrix\n" "where \"direction\" is one of\n" " row - normalize rows to one\n" " column - normalize columns to one\n" "and \"how\" is one of\n" " sum - sum adds to one after normalization\n" " length - Euclidian length as a vector adds to one\n" "options:\n" " -target=val - use target val instead of one for normalizing\n" - " -threads=N - number of threads to use, Expect limited help if you increase this\n" - " Default is %d. Only goes multitheaded in column mode and even then\n" - " about 60%% of program is in the non-parallel bit. Still, it helps\n" - , clThreads + " -serial - disable use of pthreads to speed up via parallelization\n" ); } /* Command line validation table. */ static struct optionSpec options[] = { {"target", OPTION_DOUBLE}, - {"threads", OPTION_INT}, + {"serial", OPTION_BOOLEAN}, {NULL, 0}, }; boolean howIsLength(char *how) /* Return TRUE if how is "length", FALSE if it is "sum", and abort otherwise */ { if (sameWord(how, "length")) return TRUE; else if (sameWord(how, "sum")) return FALSE; errAbort("Unrecognized \"how\" %s", how); return FALSE; } void matrixNormalizeRows(char *inFile, boolean isLength, double target, char *outFile) @@ -82,62 +77,30 @@ fprintf(f, "%s", label); for (i=0; icenterLabel); -int x, xSize = m->xSize; -for (x=0; xxLabels[x]); -fprintf(f, "\n"); - -/* Print rest */ -/* Now output going through row by row */ -int y, ySize = m->ySize; -for (y=0; yyLabels[y]); - double *row = m->rows[y]; - for (x=0; xySize; int y; if (isLength) { for (y=0; yrows[y][x]; sum += val*val; } @@ -163,77 +126,81 @@ { struct columnJob *next; int x; /* Index of column withing matrix */ boolean isLength; /* Do sqrt(sum(squares)) type normalization */ double target; /* Want column to end up summing to this */ }; void columnWorker(void *item, void *context) /* A worker to execute a single column job */ { struct columnJob *job = item; struct memMatrix *m = context; normCol(m, job->x, job->target, job->isLength); } -void matrixNormalizeColumns(char *inFile, boolean isLength, double target, char *outFile) +void matrixNormalizeColumns(char *inFile, boolean isLength, double target, char *outFile, + int threads) /* Normalize matrix one row at a time. */ { /* Open up input file */ +verboseTimeInit(); struct memMatrix *m = memMatrixFromTsv(inFile); +verboseTime(1, "Loaded %d x %d matrix from %s", m->xSize, m->ySize, inFile); /* Go through matrix by column (y-dimension) */ int xSize = m->xSize; int x; -if (clThreads > 1) +if (threads > 1) { - int threadCount = min(clThreads, 100); struct columnJob *jobList = NULL; for (x=0; xx = x; job->isLength = isLength; job->target = target; slAddHead(&jobList, job); } slReverse(&jobList); - pthreadDoList(threadCount, jobList, columnWorker, m); + pthreadDoList(threads, jobList, columnWorker, m); } else { for (x=0; x