diff options
| -rw-r--r-- | Makefile | 6 | ||||
| -rw-r--r-- | redist/dclapack.h | 2 | ||||
| -rw-r--r-- | redist/test_dcl.c | 110 |
3 files changed, 57 insertions, 61 deletions
@@ -3,6 +3,7 @@ all : lib data_recorder test calibrate calibrate_client simple_pose_test CC?=gcc CFLAGS:=-Iinclude/libsurvive -fPIC -g -O0 -Iredist -flto -DUSE_DOUBLE -std=gnu99 -rdynamic -llapacke -lcblas -lm +CFLAGS_RELEASE:=-Iinclude/libsurvive -fPIC -msse2 -ftree-vectorize -O3 -Iredist -flto -DUSE_DOUBLE -std=gnu99 -rdynamic -llapacke -lcblas -lm #LDFLAGS:=-L/usr/local/lib -lpthread -lusb-1.0 -lz -lm -flto -g LDFLAGS:=-L/usr/local/lib -lpthread -lz -lm -flto -g @@ -88,7 +89,10 @@ static_calibrate : calibrate.c redist/os_generic.c $(DRAWFUNCTIONS) $(LIBSURVIVE sed -i 's/#/\/\/#/g' ./redist/dclhelpers_debuggable.c -test_dcl: ./redist/test_dcl.c ./redist/dclhelpers_debuggable.c ./redist/dclhelpers.h ./redist/dclapack.h +test_dcl: ./redist/test_dcl.c ./redist/dclhelpers.c ./redist/dclhelpers.h ./redist/dclapack.h redist/os_generic.c + $(CC) -o $@ $^ $(LDFLAGS) $(CFLAGS_RELEASE) -DFLT=double + +test_dcl_debug: ./redist/test_dcl.c ./redist/dclhelpers_debuggable.c ./redist/dclhelpers.h ./redist/dclapack.h redist/os_generic.c $(CC) -o $@ $^ $(LDFLAGS) $(CFLAGS) -DFLT=double -fsanitize=address -fsanitize=undefined test_minimal_cv: ./src/epnp/test_minimal_cv.c ./lib/libsurvive.so diff --git a/redist/dclapack.h b/redist/dclapack.h index d4634ac..e43a4f9 100644 --- a/redist/dclapack.h +++ b/redist/dclapack.h @@ -27,7 +27,7 @@ printf(#A "\n"); \ for (int _i = 0; _i < (m); _i++) { \ for (int _j = 0; _j < (n); _j++) { \ - printf("%4.3f\t", _(A, _i, _j)); \ + printf("%4.3f ", _(A, _i, _j)); \ } \ printf("\n"); \ } \ diff --git a/redist/test_dcl.c b/redist/test_dcl.c index 5435a3c..68a6129 100644 --- a/redist/test_dcl.c +++ b/redist/test_dcl.c @@ -1,13 +1,60 @@ //gcc -msse2 -O3 -ftree-vectorize test_dcl.c dclhelpers.c os_generic.c -DFLT=double -lpthread -lcblas && valgrind ./a.out - #include "dclhelpers.h" +#include "os_generic.h" #include <assert.h> +#include <cblas.h> #include <math.h> #include <stdint.h> #include <stdio.h> -#include "os_generic.h" -#include <cblas.h> +#include <stdlib.h> + +void compareToCblas() { + FLT em1[12][20]; + FLT em2[20][20]; + FLT emo[2][20][20] = {}; + int x, y; + + for (y = 0; y < 12; y++) + for (x = 0; x < 20; x++) + em1[y][x] = (rand() % 1000) / 1000.0; + + for (y = 0; y < 20; y++) + for (x = 0; x < 20; x++) + em2[y][x] = (rand() % 1000) / 1000.0; + + int m = 12; + int n = 20; + int k = 20; + + dclPrint(DMS(em1), 12, 20); + dclPrint(DMS(em2), 20, 12); + + double times[2]; + for (int z = 0; z < 2; z++) { + double start = OGGetAbsoluteTime(); + for (int i = 0; i < 100000; i++) { + dclZero(DMS(emo[z]), 20, 20); + + if (z == 0) { + dcldgemm(0, 0, m, n, k, 1.0, DMS(em1), DMS(em2), .1, DMS(emo[z])); + } else { + cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, m, n, k, 1.0, DMS(em1), DMS(em2), .1, + DMS(emo[z])); + } + /*void cblas_dgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA, + CBLAS_TRANSPOSE TransB, const int M, const int N, + const int K, const double alpha, const double *A, + const int lda, const double *B, const int ldb, + const double beta, double *C, const int ldc);*/ + } + + printf("%s Elapsed: %f\n", z ? "CBlas" : "dcl", times[z] = OGGetAbsoluteTime() - start); + } + printf("%fx difference\n", times[0] / times[1]); + dclPrint(emo[0][0], 12, 20, 12); + dclPrint(emo[1][0], 12, 20, 12); +} int main() { @@ -65,61 +112,6 @@ int main() } } - -#if 1 - - //Currently failing test... - { -// FLT em1[3][4]; -// FLT em2[4][2]; -// FLT emo[4][2]; - - FLT em1[12][20]; - FLT em2[20][20]; - FLT emo[20][20]; - int x, y; - - for( y = 0; y < 12; y++ ) - for( x = 0; x < 20; x++ ) - em1[y][x] = (rand()%1000)/1000.0; - - for( y = 0; y < 20; y++ ) - for( x = 0; x < 20; x++ ) - em2[y][x] = (rand()%1000)/1000.0; - - for( y = 0; y < 20; y++ ) - for( x = 0; x < 20; x++ ) - emo[y][x] = 0; - - int m = 12; - int n = 20; - int k = 12; - - dclPrint( DMS(em1), 12, 20 ); - dclPrint( DMS(em2), 20, 12 ); - - int i; - - double start = OGGetAbsoluteTime(); - for( i = 0; i < 10000; i++ ) - { - dclZero( DMS(emo), 20, 20 ); - - dcldgemm( 0, 0, m, n, k, 1.0, DMS(em1), DMS(em2), .1, DMS(emo) ); - //cblas_dgemm( CblasColMajor, CblasNoTrans, CblasNoTrans, m, n, k, 1.0, DMS(em1), DMS(em2), .1, DMS(emo) ); - -/*void cblas_dgemm(CBLAS_LAYOUT layout, CBLAS_TRANSPOSE TransA, - CBLAS_TRANSPOSE TransB, const int M, const int N, - const int K, const double alpha, const double *A, - const int lda, const double *B, const int ldb, - const double beta, double *C, const int ldc);*/ - - } - printf( "Elapsed: %f\n", OGGetAbsoluteTime()-start ); - - dclPrint( emo[0], 12, 20, 12 ); - } -#endif - + compareToCblas(); } |