Merge branch 'master' of gitorious.design.ti.com:linearalgebra/linalg

diff --git a/Makefile b/Makefile
index 76503d78ec1bbd6ccbf3b77469d97894621a221f..363c44d4ccef0a821035545c131a450500b730ea 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -42,5 +42,4 @@ install:
        cp $(LINALG_CBLAS_DIR)/include/cblas.h ${DESTDIR}/usr/include
        cp $(LINALG_BLISACC_DIR)/lib/libcblas_armplusdsp.a ${DESTDIR}/usr/lib
        cp $(LINALG_BLIS_DIR)/install/arm/lib/libblis-*-cortex-a15.a ${DESTDIR}/usr/lib/libblis.a
-
-       
\ No newline at end of file
+       cp ./examples ${DESTDIR}/usr/share/ti/examples/linalg

diff --git a/build/tar_files_list.txt b/build/tar_files_list.txt
index 8cb3afdcd32f150e9cc58d691e8f0b5d8715cd56..588088c19a632493efd978f6b5fe1688f3e823be 100644 (file)
--- a/build/tar_files_list.txt
+++ b/build/tar_files_list.txt
@@ -1,6 +1,7 @@
 make.inc
 Makefile
 debian
+examples
 blis/build
 blis/CHANGELOG
 blis/config

diff --git a/examples/Makefile b/examples/Makefile
new file mode 100644 (file)
index 0000000..7152704
--- /dev/null
+++ b/examples/Makefile
@@ -0,0 +1,28 @@
+.SILENT:
+
+MFS  = $(wildcard */Makefile)
+DIRS = $(patsubst %/Makefile,%,$(MFS))
+
+all:
+       for dir in $(DIRS); do \
+         echo "=============== " $$dir " =================" ; \
+         $(MAKE) -C $$dir; \
+       done
+
+test:
+       for dir in $(DIRS); do \
+         echo "=============== " $$dir " =================" ; \
+          $(MAKE) -C $$dir test; \
+        done
+
+cross:
+       for dir in $(DIRS); do \
+         echo "=============== " $$dir " =================" ; \
+          $(MAKE) -C $$dir cross; \
+        done
+
+clean:
+       for dir in $(DIRS); do \
+         $(MAKE) -C $$dir clean; \
+       done
+

diff --git a/examples/make.inc b/examples/make.inc
new file mode 100644 (file)
index 0000000..0125a07
--- /dev/null
+++ b/examples/make.inc
@@ -0,0 +1,24 @@
+
+
+CC = gcc
+CFLAGS = -g -O2 -I/usr/include
+
+BLAS_LIB_DIR = /usr/lib/
+BLASLIB = $(BLAS_LIB_DIR)libcblas_armplusdsp.a $(BLAS_LIB_DIR)libblis.a -lOpenCL -locl_util -lstdc++ -lrt -lm -lgomp
+
+
+%.o: %.c
+       @$(CC) -c $(CFLAGS) $<
+       @echo Compiling $<
+
+$(EXE):
+
+cross: $(EXE)
+
+clean::
+       @rm -f $(EXE) *.o *.obj *.out *.asm *.if *.opt *.bc *.objc *.map *.bin *.dsp_h
+
+test: clean $(EXE)
+       @echo Running   $(EXE)
+       @./$(EXE) >> /dev/null
+       @if [ $$? -ne 0 ] ; then echo "FAILED !!!" ; fi

diff --git a/examples/matmpy/Makefile b/examples/matmpy/Makefile
new file mode 100644 (file)
index 0000000..46f74fb
--- /dev/null
+++ b/examples/matmpy/Makefile
@@ -0,0 +1,8 @@
+
+EXE = matmpy
+
+include ../make.inc
+
+$(EXE): main.o
+       $(CC) $(CFLAGS) main.o $(BLASLIB) -o $@
+

diff --git a/examples/matmpy/main.c b/examples/matmpy/main.c
new file mode 100644 (file)
index 0000000..4298c08
--- /dev/null
+++ b/examples/matmpy/main.c
@@ -0,0 +1,102 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <time.h>
+
+#include "cblas.h"
+
+/* Number of elements in matrix to display */
+
+/* For profiling */
+#define tick()  clock_gettime(CLOCK_MONOTONIC, &t0);
+#define tock() (clock_gettime(CLOCK_MONOTONIC, &t1), \
+                        t1.tv_sec - t0.tv_sec + (t1.tv_nsec - t0.tv_nsec) / 1e9)
+#define fout stdout
+
+
+double *A, *B, *C;
+int m, n, k;
+double alpha, beta;
+struct timespec t0, t1;
+double secs = 0.0;
+
+static void report_flops(double secs, int m, int n, int k, int N)
+{
+  fprintf(fout,"Total time for %d tests: %8.6fs, %5.3f Mflops\n",
+          N, secs, (float)N*m*n*(2*k-1) / (secs * 1e6));
+}
+
+double matrix_mult(void) {
+    int i,j;
+    for (i = 0; i < (m*k); i++) {
+        A[i] = (double)rand()/RAND_MAX;
+    }
+
+    for (i = 0; i < (k*n); i++) {
+        B[i] = (double)rand()/RAND_MAX;
+    }
+
+    for (i = 0; i < (m*n); i++) {
+        C[i] = 0.0;
+    }
+
+    tick();
+    cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, m, n, k, alpha, A, k, B, n, beta, C, n);
+    secs += tock();
+
+    /* We do a simplistic checksum across a subset of the result matrix */
+    double checksum = 0.0;
+    for (i=0; i<m; i++) 
+      for (j=0; j<n; j++) 
+        checksum += C[j+i*n];
+    return checksum;
+}
+
+int main()
+{
+    int t;
+       double checksum;
+       int numtests = 10;
+
+    /* configuration */
+    m = k = n = 1000;
+    alpha = 0.7; 
+       beta  = 1.3; 
+
+    /* allocate the matrices */
+    A = (double *)malloc( m*k*sizeof( double ) );
+    B = (double *)malloc( k*n*sizeof( double ) );
+    C = (double *)malloc( m*n*sizeof( double ) );
+    if (A == NULL || B == NULL || C == NULL) {
+      printf( "\nERROR: Can't allocate memory for matrices. Aborting... \n\n");
+      free(A);
+      free(B);
+      free(C);
+      return 1;
+    }
+       
+    srand(123456789);
+
+    printf ("A(%ix%i) X B(%ix%i) => C(%ix%i)\n", m, k, k, n, m, n);
+
+    printf("Warming caches (by doing a single matrix-multiply)..\n");
+    checksum = matrix_mult();
+
+    /* reset secs, so we can now begin the real timing */
+    secs = 0;
+
+    printf("Now doing %d tests after warming caches\n", numtests);
+    for (t=0; t<numtests; t++)
+      checksum += matrix_mult();
+    report_flops(secs, m, n, k, numtests);
+
+    printf("Result CHECKSUM: %16.4f\n", checksum);
+
+    free(A);
+    free(B);
+    free(C);
+
+    return 0;
+}
+
+