fixes #32 autoscale based on CPUs available

This should work on both Windows and the most common POSIX
variants.  We will create at least two threads for running
completions, but there are numerous other threads in the code.

4 files changed, 32 insertions, 4 deletions

diff --git a/src/core/platform.h b/src/core/platform.h
index 07bfc14c..b709e3ba 100644
--- a/src/core/platform.h
+++ b/src/core/platform.h
@@ -190,6 +190,10 @@ extern int nni_plat_init(int (*)(void));
 // will be called until nni_platform_init is called.
 extern void nni_plat_fini(void);
 
+// nni_plat_ncpu returns the number of logical CPUs on the system.  This is
+// used to scale the number of independent threads started.
+extern int nni_plat_ncpu(void);
+
 //
 // TCP Support.
 //
diff --git a/src/core/taskq.c b/src/core/taskq.c
index ae66ec67..82a7456b 100644
--- a/src/core/taskq.c
+++ b/src/core/taskq.c
@@ -261,11 +261,14 @@ nni_task_fini(nni_task *task)
 int
 nni_taskq_sys_init(void)
 {
-	int rv;
+	int nthrs;
 
-	// XXX: Make the "16" = NCPUs * 2
-	rv = nni_taskq_init(&nni_taskq_systq, 16);
-	return (rv);
+	nthrs = nni_plat_ncpu() * 2;
+	if (nthrs < 2) {
+		nthrs = 2;
+	}
+
+	return (nni_taskq_init(&nni_taskq_systq, nthrs));
 }
 
 void
diff --git a/src/platform/posix/posix_thread.c b/src/platform/posix/posix_thread.c
index f016ca3f..df2ee9d2 100644
--- a/src/platform/posix/posix_thread.c
+++ b/src/platform/posix/posix_thread.c
@@ -336,4 +336,16 @@ nni_plat_fini(void)
 	pthread_mutex_unlock(&nni_plat_init_lock);
 }
 
+int
+nni_plat_ncpu(void)
+{
+	// POSIX specifies sysconf exists, but not the value
+	// _SC_NPROCESSORS_ONLN.  Nonetheless, everybody implements it.
+	// If you don't we'll assume you only have a single logical CPU.
+#ifdef _SC_NPROCESSORS_ONLN
+	return (sysconf(_SC_NPROCESSORS_ONLN));
+#else
+	return (1);
+#endif
+}
 #endif // NNG_PLATFORM_POSIX
diff --git a/src/platform/windows/win_thread.c b/src/platform/windows/win_thread.c
index 2e9d58d7..a3d932aa 100644
--- a/src/platform/windows/win_thread.c
+++ b/src/platform/windows/win_thread.c
@@ -148,6 +148,15 @@ nni_plat_thr_is_self(nni_plat_thr *thr)
 static LONG plat_inited = 0;
 
 int
+nni_plat_ncpu(void)
+{
+	SYSTEM_INFO info;
+
+	GetSystemInfo(&info);
+	return ((int) (info.dwNumberOfProcessors));
+}
+
+int
 nni_plat_init(int (*helper)(void))
 {
 	int            rv   = 0;