Fit crash with FFT benchmark.

8 files changed, 83 insertions, 43 deletions

diff --git a/hardinfo2/arch/common/blowfish.h b/hardinfo2/arch/common/blowfish.h
index 5fea2e22..333fb157 100644
--- a/hardinfo2/arch/common/blowfish.h
+++ b/hardinfo2/arch/common/blowfish.h
@@ -19,7 +19,7 @@
 #include <blowfish.h>
 
 static gpointer
-parallel_blowfish(unsigned int start, unsigned int end, void *data, GTimer *timer)
+parallel_blowfish(unsigned int start, unsigned int end, void *data, gint thread_number)
 {
     BLOWFISH_CTX ctx;
     unsigned int i;
diff --git a/hardinfo2/arch/common/cryptohash.h b/hardinfo2/arch/common/cryptohash.h
index 9897bb6b..2c917c9b 100644
--- a/hardinfo2/arch/common/cryptohash.h
+++ b/hardinfo2/arch/common/cryptohash.h
@@ -39,12 +39,12 @@ static void inline sha1_step(char *data, glong srclen)
     SHA1Final(checksum, &ctx);
 }
 
-static gpointer cryptohash_for(unsigned int start, unsigned int end, void *data, GTimer *timer)
+static gpointer cryptohash_for(unsigned int start, unsigned int end, void *data, gint thread_number)
 {
     unsigned int i;
     
     for (i = start; i <= end; i++) { 
-        if (i % 2 == 0) {
+        if (i & 1) {
             md5_step(data, 65536);
         } else {
             sha1_step(data, 65536);
diff --git a/hardinfo2/arch/common/fft.h b/hardinfo2/arch/common/fft.h
index 62daa9fe..f1a0ced8 100644
--- a/hardinfo2/arch/common/fft.h
+++ b/hardinfo2/arch/common/fft.h
@@ -18,12 +18,14 @@
 
 #include <fftbench.h>
 
-static gpointer fft_for(unsigned int start, unsigned int end, void *data, GTimer *timer)
+static gpointer fft_for(unsigned int start, unsigned int end, void *data, gint thread_number)
 {
     unsigned int i;
+    FFTBench **benches = (FFTBench **)data;
+    FFTBench *fftbench = (FFTBench *)(benches[thread_number]);
     
     for (i = start; i <= end; i++) { 
-        fft_bench_start();
+        fft_bench_run(fftbench);
     }
     
     return NULL;
@@ -33,15 +35,31 @@ static void
 benchmark_fft(void)
 {
     gdouble elapsed = 0;
+    int n_cores, i;
+    gchar *temp;
+    FFTBench **benches;
     
     shell_view_set_enabled(FALSE);
     shell_status_update("Running FFT benchmark...");
         
-    fft_bench_init();
-    elapsed = benchmark_parallel_for(0, 4, fft_for, NULL);
-    fft_bench_finish();
+    /* Pre-allocate all benchmarks */
+    temp = module_call_method("devices::getProcessorCount");
+    n_cores = temp ? atoi(temp) : 1;
+    g_free(temp);
     
+    benches = g_new0(FFTBench *, n_cores);
+    for (i = 0; i < n_cores; i++) {
+      benches[i] = fft_bench_new();
+    }
+    
+    /* Run the benchmark */
+    elapsed = benchmark_parallel_for(0, 4, fft_for, benches);
+    
+    /* Free up the memory */
+    for (i = 0; i < n_cores; i++) {
+      fft_bench_free(benches[i]);
+    }
+    g_free(benches);
+        
     bench_results[BENCHMARK_FFT] = elapsed;
 }
-
-
diff --git a/hardinfo2/arch/common/nqueens.h b/hardinfo2/arch/common/nqueens.h
index 2a233722..6ad5b4fd 100644
--- a/hardinfo2/arch/common/nqueens.h
+++ b/hardinfo2/arch/common/nqueens.h
@@ -18,7 +18,7 @@
 
 #include <nqueens.h>
 
-static gpointer nqueens_for(unsigned int start, unsigned int end, void *data, GTimer *timer)
+static gpointer nqueens_for(unsigned int start, unsigned int end, void *data, gint thread_number)
 {
     unsigned int i;
     
diff --git a/hardinfo2/arch/common/raytrace.h b/hardinfo2/arch/common/raytrace.h
index 7fdc5e21..fe2fda91 100644
--- a/hardinfo2/arch/common/raytrace.h
+++ b/hardinfo2/arch/common/raytrace.h
@@ -19,7 +19,7 @@
 void fbench();	/* fbench.c */
 
 static gpointer
-parallel_raytrace(unsigned int start, unsigned int end, gpointer data)
+parallel_raytrace(unsigned int start, unsigned int end, gpointer data, gint thread_number)
 {
     unsigned int i;
     
diff --git a/hardinfo2/benchmark.c b/hardinfo2/benchmark.c
index e31d6b34..d91394bf 100644
--- a/hardinfo2/benchmark.c
+++ b/hardinfo2/benchmark.c
@@ -62,6 +62,7 @@ static ModuleEntry entries[] = {
 typedef struct _ParallelBenchTask ParallelBenchTask;
 
 struct _ParallelBenchTask {
+    gint	thread_number;
     guint	start, end;
     gpointer	data, callback;
 };
@@ -69,13 +70,13 @@ struct _ParallelBenchTask {
 gpointer benchmark_parallel_for_dispatcher(gpointer data)
 {
     ParallelBenchTask 	*pbt = (ParallelBenchTask *)data;
-    gpointer 		(*callback)(unsigned int start, unsigned int end, void *data);
+    gpointer 		(*callback)(unsigned int start, unsigned int end, void *data, gint thread_number);
     gpointer		return_value;
     
     if ((callback = pbt->callback)) {
         DEBUG("this is thread %p; items %d -> %d, data %p", g_thread_self(),
               pbt->start, pbt->end, pbt->data);
-        return_value = callback(pbt->start, pbt->end, pbt->data);
+        return_value = callback(pbt->start, pbt->end, pbt->data, pbt->thread_number);
         DEBUG("this is thread %p; return value is %p", g_thread_self(), return_value);
     } else {
         DEBUG("this is thread %p; callback is NULL and it should't be!", g_thread_self());
@@ -89,7 +90,7 @@ gpointer benchmark_parallel_for_dispatcher(gpointer data)
 gdouble benchmark_parallel_for(guint start, guint end,
                                gpointer callback, gpointer callback_data) {
     gchar	*temp;
-    guint	n_cores, iter_per_core, iter;
+    guint	n_cores, iter_per_core, iter, thread_number = 0;
     gdouble	elapsed_time;
     GSList	*threads = NULL, *t;
     GTimer	*timer;
@@ -121,7 +122,8 @@ gdouble benchmark_parallel_for(guint start, guint end,
         
         DEBUG("launching thread %d", 1 + (iter / iter_per_core));
         
-        pbt->start    = iter == 0 ? 0 : iter + 1;
+        pbt->thread_number = thread_number++;
+        pbt->start    = iter == 0 ? 0 : iter;
         pbt->end      = iter + iter_per_core - 1;
         pbt->data     = callback_data;
         pbt->callback = callback;
diff --git a/hardinfo2/fftbench.c b/hardinfo2/fftbench.c
index 597c5693..d51dbb1d 100644
--- a/hardinfo2/fftbench.c
+++ b/hardinfo2/fftbench.c
@@ -30,6 +30,8 @@
 #include <stdbool.h>
 #include <stdio.h>
 
+#include "fftbench.h"
+
 // embedded random number generator; ala Park and Miller
 static long seed = 1325;
 static const long IA = 16807;
@@ -61,13 +63,15 @@ static const int N = 800;
 static const int NM1 = 799;	// N - 1
 static const int NP1 = 801;	// N + 1
 
-static int *lup_decompose(double **a)
+static void lup_decompose(FFTBench *fftbench)
 {
     int i, j, k, k2, t;
-    double p, temp;
+    double p, temp, **a;
 
     int *perm = (int *) malloc(sizeof(double) * N);
-
+    
+    a = fftbench->a;
+    
     for (i = 0; i < N; ++i)
 	perm[i] = i;
 
@@ -85,7 +89,7 @@ static int *lup_decompose(double **a)
 
 	// check for invalid a
 	if (p == 0.0)
-	    return NULL;
+	    return;
 
 	// exchange rows
 	t = perm[k];
@@ -106,16 +110,20 @@ static int *lup_decompose(double **a)
 	}
     }
 
-    return perm;
+    fftbench->p = perm;
 }
 
-static double *lup_solve(double **a, int *perm, double *b)
+static double *lup_solve(FFTBench *fftbench)
 {
     int i, j, j2;
     double sum, u;
 
     double *y = (double *) malloc(sizeof(double) * N);
     double *x = (double *) malloc(sizeof(double) * N);
+    
+    double **a = fftbench->a;
+    double *b = fftbench->b;
+    int *perm = fftbench->p;
 
     for (i = 0; i < N; ++i) {
 	y[i] = 0.0;
@@ -156,46 +164,49 @@ static double *lup_solve(double **a, int *perm, double *b)
     return x;
 }
 
-static double **a, *b, *r;
-static int *p;
-
-void fft_bench_init(void)
+FFTBench *fft_bench_new(void)
 {
+    FFTBench *fftbench;
     int i, j;
+    
+    fftbench = g_new0(FFTBench, 1);
 
     // generate test data            
-    a = (double **) malloc(sizeof(double *) * N);
+    fftbench->a = (double **) malloc(sizeof(double *) * N);
 
     for (i = 0; i < N; ++i) {
-	a[i] = (double *) malloc(sizeof(double) * N);
+	fftbench->a[i] = (double *) malloc(sizeof(double) * N);
 
 	for (j = 0; j < N; ++j)
-	    a[i][j] = random_double();
+	    fftbench->a[i][j] = random_double();
     }
 
-    b = (double *) malloc(sizeof(double) * N);
+    fftbench->b = (double *) malloc(sizeof(double) * N);
 
     for (i = 0; i < N; ++i)
-	b[i] = random_double();
+	fftbench->b[i] = random_double();
 
+    return fftbench;
 }
 
-void fft_bench_start(void)
+void fft_bench_run(FFTBench *fftbench)
 {
-    p = lup_decompose(a);
-    r = lup_solve(a, p, b);
+    lup_decompose(fftbench);
+    lup_solve(fftbench);
 }
 
-void fft_bench_finish(void)
+void fft_bench_free(FFTBench *fftbench)
 {
     int i;
     
     // clean up
     for (i = 0; i < N; ++i)
-	free(a[i]);
+	free(fftbench->a[i]);
 
-    free(a);
-    free(b);
-    free(p);
-    free(r);
+    free(fftbench->a);
+    free(fftbench->b);
+    free(fftbench->p);
+    free(fftbench->r);
+    
+    g_free(fftbench);
 }
diff --git a/hardinfo2/fftbench.h b/hardinfo2/fftbench.h
index 3598baf1..7c0afc3d 100644
--- a/hardinfo2/fftbench.h
+++ b/hardinfo2/fftbench.h
@@ -1,9 +1,18 @@
 #ifndef __FFTBENCH_H__ 
 #define __FFTBENCH_H__
 
-void fft_bench_init(void);
-void fft_bench_start(void);
-void fft_bench_finish(void);
+#include <glib.h>
+
+typedef struct _FFTBench	FFTBench;
+
+struct _FFTBench {
+  double **a, *b, *r;
+  int *p;
+};
+
+FFTBench *fft_bench_new(void);
+void fft_bench_run(FFTBench *fftbench);
+void fft_bench_free(FFTBench *fftbench);
 
 #endif /* __FFTBENCH_H__ */