From 29fc429d9a34a513e737df287c8a8785c9f7a332 Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Tue, 11 Jul 2017 18:27:33 +0200
Subject: [PATCH] Honor cgroup/cpuset constraints when enumerating cpus

---
 driver/others/init.c   | 59 ++++++++++++++++++++++++++++++++++++++++++---
 driver/others/memory.c | 65 ++++++++++++++++++++++++++++++++------------------
 2 files changed, 98 insertions(+), 26 deletions(-)

diff --git a/driver/others/init.c b/driver/others/init.c
index 9be6f52..4093776 100644
--- a/driver/others/init.c
+++ b/driver/others/init.c
@@ -354,6 +354,24 @@ static int numa_check(void) {
   return common -> num_nodes;
 }
 
+#if defined(__GLIBC_PREREQ)
+#if !__GLIBC_PREREQ(2, 6)
+int sched_getcpu(void)
+{
+int cpu;
+FILE *fp = NULL;
+if ( (fp = fopen("/proc/self/stat", "r")) == NULL)
+   return -1;
+if ( fscanf( fp, "%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%*s%d", &cpu) != 1) {
+  fclose (fp);
+  return -1;
+  }
+  fclose (fp);
+  return(cpu);
+}
+#endif
+#endif
+
 static void numa_mapping(void) {
 
   int node, cpu, core;
@@ -808,16 +826,51 @@ void gotoblas_affinity_init(void) {
   common -> shmid = pshmid;
 
   if (common -> magic != SH_MAGIC) {
-
+    cpu_set_t *cpusetp;
+    int nums;
+    int ret;
 #ifdef DEBUG
     fprintf(stderr, "Shared Memory Initialization.\n");
 #endif
 
     //returns the number of processors which are currently online
-    common -> num_procs = sysconf(_SC_NPROCESSORS_CONF);;
+    nums = sysconf(_SC_NPROCESSORS_CONF);
+     
+#if !defined(__GLIBC_PREREQ) || !__GLIBC_PREREQ(2, 3)
+    common->num_procs = nums;
+#elif __GLIBC_PREREQ(2, 7)
+    cpusetp = CPU_ALLOC(nums);
+    if (cpusetp == NULL) {
+        common->num_procs = nums;
+    } else {
+        size_t size;
+        size = CPU_ALLOC_SIZE(nums);
+        ret = sched_getaffinity(0,size,cpusetp);
+        if (ret!=0) 
+            common->num_procs = nums;
+        else
+            common->num_procs = CPU_COUNT_S(size,cpusetp);
+    }
+    CPU_FREE(cpusetp);
+#else
+    ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
+    if (ret!=0) { 
+        common->num_procs = nums;
+    } else {
+#if !__GLIBC_PREREQ(2, 6)  
+    int i;
+    int n = 0;
+    for (i=0;i<nums;i++)
+        if (CPU_ISSET(i,cpusetp)) n++;
+    common->num_procs = n;   
+#else
+    common->num_procs = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
+#endif
+
+#endif 
 
     if(common -> num_procs > MAX_CPUS) {
-      fprintf(stderr, "\nOpenBLAS Warining : The number of CPU/Cores(%d) is beyond the limit(%d). Terminated.\n", common->num_procs, MAX_CPUS);
+      fprintf(stderr, "\nOpenBLAS Warning : The number of CPU/Cores(%d) is beyond the limit(%d). Terminated.\n", common->num_procs, MAX_CPUS);
       exit(1);
     }
 
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 9169503..103c970 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -175,7 +175,44 @@ int get_num_procs(void);
 #else
 int get_num_procs(void) {
   static int nums = 0;
+cpu_set_t *cpusetp;
+size_t size;
+int ret;
+int i,n;
+
   if (!nums) nums = sysconf(_SC_NPROCESSORS_CONF);
+#if !defined(OS_LINUX)
+     return nums;
+#endif
+     
+#if !defined(__GLIBC_PREREQ)
+   return nums;
+#endif   
+#if !__GLIBC_PREREQ(2, 3)
+   return nums;
+#endif   
+
+#if !__GLIBC_PREREQ(2, 7)
+  ret = sched_getaffinity(0,sizeof(cpu_set_t), cpusetp);
+  if (ret!=0) return nums;
+  n=0;
+#if !__GLIBC_PREREQ(2, 6)  
+  for (i=0;i<nums;i++)
+     if (CPU_ISSET(i,cpusetp)) n++;
+  nums=n;   
+#else
+  nums = CPU_COUNT(sizeof(cpu_set_t),cpusetp);
+#endif
+  return nums;
+#endif
+
+  cpusetp = CPU_ALLOC(nums);
+  if (cpusetp == NULL) return nums;
+  size = CPU_ALLOC_SIZE(nums);
+  ret = sched_getaffinity(0,size,cpusetp);
+  if (ret!=0) return nums;
+  nums = CPU_COUNT_S(size,cpusetp);
+  CPU_FREE(cpusetp);
   return nums;
 }
 #endif
@@ -1015,7 +1052,7 @@ void *blas_memory_alloc(int procpos){
   mypos = WhereAmI();
 
   position = mypos;
-  while (position >= NUM_BUFFERS) position >>= 1;
+  while (position > NUM_BUFFERS) position >>= 1;
 
   do {
     if (!memory[position].used && (memory[position].pos == mypos)) {
@@ -1164,8 +1201,8 @@ void blas_memory_free(void *free_area){
   position = 0;
   LOCK_COMMAND(&alloc_lock);
 
-  while ((position < NUM_BUFFERS) && (memory[position].addr != free_area))
-    position++;
+  while ((memory[position].addr != free_area)
+	 && (position < NUM_BUFFERS)) position++;
 
   if (memory[position].addr != free_area) goto error;
 
@@ -1479,30 +1516,12 @@ static int on_process_term(void)
 #else
 #pragma comment(linker, "/INCLUDE:__tls_used")
 #endif
-
-#ifdef _WIN64
-#pragma const_seg(".CRT$XLB")
-#else
+#pragma data_seg(push, old_seg)
 #pragma data_seg(".CRT$XLB")
-#endif
 static void (APIENTRY *dll_callback)(HINSTANCE h, DWORD ul_reason_for_call, PVOID pv) = DllMain;
-#ifdef _WIN64
-#pragma const_seg()
-#else
-#pragma data_seg()
-#endif
-
-#ifdef _WIN64
-#pragma const_seg(".CRT$XTU")
-#else
 #pragma data_seg(".CRT$XTU")
-#endif
 static int(*p_process_term)(void) = on_process_term;
-#ifdef _WIN64
-#pragma const_seg()
-#else
-#pragma data_seg()
-#endif
+#pragma data_seg(pop, old_seg)
 #endif
 
 #if (defined(C_PGI) || (!defined(C_SUN) && defined(F_INTERFACE_SUN))) && (defined(ARCH_X86) || defined(ARCH_X86_64))
-- 
2.7.4