fftw plan creation is not thread safe

alaeri · Mar 13, 2014 · 07c95b9 · 07c95b9
1 parent ac54d1d
commit 07c95b9
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 4 deletions.
diff --git a/bin/image-net.c b/bin/image-net.c
@@ -409,10 +409,10 @@ int main(int argc, char** argv)
 	for (i = 0; i < 13; i++)
 	{
 		layer_params[i].w.decay = 0.0005;
-		layer_params[i].w.learn_rate = 0.01;
+		layer_params[i].w.learn_rate = 0.001;
 		layer_params[i].w.momentum = 0.9;
 		layer_params[i].bias.decay = 0;
-		layer_params[i].bias.learn_rate = 0.01;
+		layer_params[i].bias.learn_rate = 0.001;
 		layer_params[i].bias.momentum = 0.9;
 	}
 	layer_params[10].dor = 0.5;

diff --git a/lib/ccv_numeric.c b/lib/ccv_numeric.c
@@ -2,6 +2,7 @@
 #include "ccv_internal.h"
 #include <complex.h>
 #ifdef HAVE_FFTW3
+#include <pthread.h>
 #include <fftw3.h>
 #else
 #include "3rdparty/kissfft/kiss_fftndr.h"
@@ -570,6 +571,8 @@ static int _ccv_get_optimal_fft_size(int size)
     return _ccv_optimal_fft_size[b];
 }
 
+static pthread_mutex_t fftw_plan_mutex = PTHREAD_MUTEX_INITIALIZER;
+
 static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_dense_matrix_t* d, int padding_pattern)
 {
 	int ch = CCV_GET_CHANNEL(a->type);
@@ -587,6 +590,7 @@ static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_d
 		fftw_a = fftwf_malloc(rows * cols_2c * ch * sizeof(float));
 		fftw_b = fftwf_malloc(rows * cols_2c * ch * sizeof(float));
 		fftw_d = fftwf_malloc(rows * cols_2c * ch * sizeof(float));
+		pthread_mutex_lock(&fftw_plan_mutex);
 		if (ch == 1)
 		{
 			pf = fftwf_plan_dft_r2c_2d(rows, cols, 0, 0, FFTW_ESTIMATE);
@@ -596,10 +600,12 @@ static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_d
 			pf = fftwf_plan_many_dft_r2c(2, ndim, ch, 0, 0, ch, 1, 0, 0, ch, 1, FFTW_ESTIMATE);
 			pinvf = fftwf_plan_many_dft_c2r(2, ndim, ch, 0, 0, ch, 1, 0, 0, ch, 1, FFTW_ESTIMATE);
 		}
+		pthread_mutex_unlock(&fftw_plan_mutex);
 	} else {
 		fftw_a = fftw_malloc(rows * cols_2c * ch * sizeof(double));
 		fftw_b = fftw_malloc(rows * cols_2c * ch * sizeof(double));
 		fftw_d = fftw_malloc(rows * cols_2c * ch * sizeof(double));
+		pthread_mutex_lock(&fftw_plan_mutex);
 		if (ch == 1)
 		{
 			p = fftw_plan_dft_r2c_2d(rows, cols, 0, 0, FFTW_ESTIMATE);
@@ -609,6 +615,7 @@ static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_d
 			p = fftw_plan_many_dft_r2c(2, ndim, ch, 0, 0, ch, 1, 0, 0, ch, 1, FFTW_ESTIMATE);
 			pinv = fftw_plan_many_dft_c2r(2, ndim, ch, 0, 0, ch, 1, 0, 0, ch, 1, FFTW_ESTIMATE);
 		}
+		pthread_mutex_unlock(&fftw_plan_mutex);
 	}
 	memset(fftw_b, 0, rows * cols_2c * ch * CCV_GET_DATA_TYPE_SIZE(fft_type));
 
@@ -728,8 +735,10 @@ static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_d
 		ccv_matrix_setter(d->type, ccv_matrix_getter, a->type, for_block, float, fftwf_complex);
 #undef fft_execute_dft_r2c
 #undef fft_execute_dft_c2r
+		pthread_mutex_lock(&fftw_plan_mutex);
 		fftwf_destroy_plan(pf);
 		fftwf_destroy_plan(pinvf);
+		pthread_mutex_unlock(&fftw_plan_mutex);
 		fftwf_free(fftw_a);
 		fftwf_free(fftw_b);
 		fftwf_free(fftw_d);
@@ -739,8 +748,10 @@ static void _ccv_filter_fftw(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, ccv_d
 		ccv_matrix_setter(d->type, ccv_matrix_getter, a->type, for_block, double, fftw_complex);
 #undef fft_execute_dft_r2c
 #undef fft_execute_dft_c2r
+		pthread_mutex_lock(&fftw_plan_mutex);
 		fftw_destroy_plan(p);
 		fftw_destroy_plan(pinv);
+		pthread_mutex_unlock(&fftw_plan_mutex);
 		fftw_free(fftw_a);
 		fftw_free(fftw_b);
 		fftw_free(fftw_d);

diff --git a/lib/configure b/lib/configure
@@ -3323,7 +3323,7 @@ fi
 ac_fn_c_check_header_mongrel "$LINENO" "fftw3.h" "ac_cv_header_fftw3_h" "$ac_includes_default"
 if test "x$ac_cv_header_fftw3_h" = xyes; then :
   DEFINE_MACROS="$DEFINE_MACROS-D HAVE_FFTW3 "
- MKLDFLAGS="$MKLDFLAGS-lfftw3 -lfftw3f "
+ MKLDFLAGS="$MKLDFLAGS-lfftw3 -lfftw3f -lpthread "
 
 fi
 

diff --git a/lib/configure.ac b/lib/configure.ac
@@ -19,7 +19,7 @@ AC_CHECK_HEADER(png.h,
 AC_CHECK_HEADER(jpeglib.h,
 				[AC_SUBST(DEFINE_MACROS, ["$DEFINE_MACROS-D HAVE_LIBJPEG "]) AC_SUBST(MKLDFLAGS, ["$MKLDFLAGS-ljpeg "])])
 AC_CHECK_HEADER(fftw3.h,
-				[AC_SUBST(DEFINE_MACROS, ["$DEFINE_MACROS-D HAVE_FFTW3 "]) AC_SUBST(MKLDFLAGS, ["$MKLDFLAGS-lfftw3 -lfftw3f "])])
+				[AC_SUBST(DEFINE_MACROS, ["$DEFINE_MACROS-D HAVE_FFTW3 "]) AC_SUBST(MKLDFLAGS, ["$MKLDFLAGS-lfftw3 -lfftw3f -lpthread "])])
 AC_CHECK_HEADER(linear.h,
 				[AC_SUBST(DEFINE_MACROS, ["$DEFINE_MACROS-D HAVE_LIBLINEAR "]) AC_SUBST(MKLDFLAGS, ["$MKLDFLAGS-llinear "])])
 AC_CHECK_HEADER(cblas.h,