OpenStream Options: option on/off consistency

Turn option off by settint the value to zero and on to set to any value
but zero.

Sount default options.

Author: Syllo

libworkstream_df/alloc.h

@@ -47,7 +47,7 @@ static inline int slab_force_advise_pages(void* addr, size_t size, int advice)
 	       round_page_size(size),
 	       advice))
     {
-#ifdef SLAB_ALLOCATOR_VERBOSE
+#if SLAB_ALLOCATOR_VERBOSE
 	    fprintf(stderr, "Could not disable use of huge pages\n");
 	    perror("madvise");
 #endif // SLAB_ALLOCATOR_VERBOSE
@@ -72,7 +72,7 @@ static inline int slab_get_numa_node(void* address, unsigned int size)
   hwloc_bitmap_t numa_nodes = numa_memlocation_of_memory(address, size);
   // The memory could be allocated on more than one node, return one of them
   int max_node = hwloc_bitmap_first(numa_nodes);
-#ifdef SLAB_ALLOCATOR_VERBOSE
+#if SLAB_ALLOCATOR_VERBOSE
 	if(max_node < 0)
 	  fprintf(stderr, "Could not determine node of %p\n", address);
 #endif // SLAB_ALLOCATOR_VERBOSE
@@ -229,7 +229,7 @@ static inline int slab_alloc_memalign(slab_cache_p slab_cache, void** ptr, size_
   pthread_spin_lock(&slab_cache->free_mem_lock);
 
   if(slab_cache->free_mem_bytes < alloc_size) {
-#ifdef SLAB_ALLOCATOR_VERBOSE
+#if SLAB_ALLOCATOR_VERBOSE
     if(slab_cache->free_mem_bytes)
       printf("wasted %zu bytes\n", slab_cache->free_mem_bytes);
 #endif // SLAB_ALLOCATOR_VERBOSE
@@ -328,7 +328,7 @@ slab_warmup (slab_cache_p slab_cache, unsigned int idx, unsigned int num_slabs,
   assert(!posix_memalign_success);
 
   if (bind_memory_to_numa_node(alloc, alloc_size, node)) {
-#ifdef SLAB_ALLOCATOR_VERBOSE
+#if SLAB_ALLOCATOR_VERBOSE
     fprintf(stderr, "Could not slab memory to numa node %u\n", node);
 #endif // SLAB_ALLOCATOR_VERBOSE
   }

libworkstream_df/config.h

@@ -19,10 +19,11 @@
  * When a worker did not successfully steal some work from any other workers,
  * the worker will relinquish the CPU and the thread is placed at the end of
  * the scheduler queue.
- * Activating this option will activate active polling for new task without stealing the 
+ * With this option active, workers will actively poll for new task without
+ * relinquishing the CPU.
  */
 
-// #define WS_NO_YIELD_SPIN
+#define WS_NO_YIELD_SPIN 0
 
 /*
  * Disable workers local cache. A Task placed into this local cache cannot be
@@ -67,15 +68,15 @@
  * This may have a performance impact.
  */
 
-// #define SLAB_ALLOCATOR_VERBOSE
+#define SLAB_ALLOCATOR_VERBOSE 0
 
 /*
  * Make HWLOC print information about the hardware and worker placement. This
  * has no performance impact on the program other than the initialisation in
  * the pre_main function.
  */
 
-// #define HWLOC_VERBOSE
+#define HWLOC_VERBOSE 0
 
  /*********************** OpenStream Profiling Options ***********************/
 
@@ -92,7 +93,8 @@
  * REQUIRES WQUEUE_PROFILE
  */
 
-// #define MATRIX_PROFILE "wqueue_matrix.out"
+#define MATRIX_PROFILE 0
+#define MATRIX_PROFILE_OUTPUT "wqueue_matrix.out"
 
 /*
  * Use linux getrusage function to gather resource usage for running threads.
@@ -104,7 +106,7 @@
  *   - The number of involuntary context switches (e.g. kernel scheduler intervention)
  */
 
-// #define PROFILE_RUSAGE
+#define PROFILE_RUSAGE 0
 
  /*********************** OpenStream Probably Broken Options ***********************/
 
@@ -154,12 +156,8 @@
 /*
  * Some configuration checks
  */
-#if defined(UNIFORM_MEMORY_ACCESS) && MAX_NUMA_NODES != 1
-#error "UNIFORM_MEMORY_ACCESS defined, but MAX_NUMA_NODES != 1"
-#endif
-
-#if defined(MATRIX_PROFILE) && !WQUEUE_PROFILE
+#if MATRIX_PROFILE && !WQUEUE_PROFILE
 #error "MATRIX_PROFILE defined, but WQUEUE_PROFILE != 1"
-#endif // defined(MATRIX_PROFILE) && !defined(WQUEUE_PROFILE)
+#endif
 
 #endif

libworkstream_df/hwloc-support.c

@@ -69,7 +69,7 @@ bool discover_machine_topology(void) {
       malloc(nproc * sizeof(*cpuid_to_closest_numa_node));
   populate_closest_numa_nodes();
 
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
   fprintf(stdout,
           "[HWLOC Info] The machine has a depth of %d\n"
           "[HWLOC Info] The machine has %d numa node(s)\n"
@@ -125,7 +125,7 @@ bool distribute_worker_on_topology(unsigned num_workers,
     hwloc_bitmap_or(restricted_set, restricted_set, distrib_sets[i]);
     (*processing_units)[i] = hwloc_get_next_obj_inside_cpuset_by_type(
         machine_topology, distrib_sets[i], HWLOC_OBJ_PU, NULL);
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
     fprintf(stderr, "Worker %u mapped to processing unit %u (OS index %u)\n", i,
             (*processing_units)[i]->logical_index,
             (*processing_units)[i]->os_index);

libworkstream_df/numa.c

@@ -14,7 +14,7 @@ static inline void numa_node_init(wstream_df_numa_node_p node, int node_id) {
   if (bind_memory_to_numa_node(node->workers,
                                wstream_num_workers * sizeof(*node->workers),
                                node_id)) {
-#ifdef HWLOC_VERBOSE 
+#if HWLOC_VERBOSE 
     fprintf(stderr, "Could not bind memory to numa node %u\n", node_id);
 #endif // HWLOC_VERBOSE
   }
@@ -30,7 +30,7 @@ int numa_nodes_init(void)
       wstream_df_fatal("Cannot allocate numa node structure");
 
     if (bind_memory_to_numa_node(ptr, size, i)) {
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
       fprintf(stderr, "Could not bind memory to numa node %u\n", i);
 #endif // HWLOC_VERBOSE
     }

libworkstream_df/profiling.c

@@ -6,12 +6,12 @@
 #include "numa.h"
 #include <pthread.h>
 
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
 #include <sys/time.h>
 #include <sys/resource.h>
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
-#ifdef MATRIX_PROFILE
+#if MATRIX_PROFILE
 #include <assert.h>
 
 void *tm_data__;
@@ -25,7 +25,7 @@ void init_transfer_matrix(void) {
 void dump_transfer_matrix(unsigned int num_workers)
 {
 	unsigned int i, j;
-	FILE* matrix_fp = fopen(MATRIX_PROFILE, "w+");
+	FILE* matrix_fp = fopen(MATRIX_PROFILE_OUTPUT, "w+");
 	assert(matrix_fp);
 
 	for (i = 0; i < num_workers; ++i) {
@@ -247,7 +247,7 @@ wqueue_counters_enter_runtime(struct wstream_df_thread* th)
 {
 }
 
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
 
 void wqueue_counters_profile_rusage(struct wstream_df_thread *th) {
   struct rusage usage;
@@ -263,15 +263,15 @@ void wqueue_counters_profile_rusage(struct wstream_df_thread *th) {
   th->inv_context_switches = usage.ru_nivcsw;
 }
 
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
 void init_wqueue_counters(wstream_df_thread_p th) {
   th->steals_owncached = 0;
   th->steals_ownqueue = 0;
   th->steals_mem = calloc(topology_depth, sizeof(*th->steals_mem));
   if (bind_memory_to_cpu_memspace(
           th->steals_mem, topology_depth * sizeof(*th->steals_mem), th->cpu)) {
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
     perror("hwloc_membind");
 #endif // HWLOC_VERBOSE
   }
@@ -284,7 +284,7 @@ void init_wqueue_counters(wstream_df_thread_p th) {
   th->bytes_mem = calloc(topology_depth, sizeof(*th->bytes_mem));
   if (bind_memory_to_cpu_memspace(
           th->bytes_mem, topology_depth * sizeof(*th->bytes_mem), th->cpu)) {
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
     perror("hwloc_membind");
 #endif // HWLOC_VERBOSE
   }
@@ -297,13 +297,13 @@ void init_wqueue_counters(wstream_df_thread_p th) {
 
   th->reuse_addr = 0;
   th->reuse_copy = 0;
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
   th->system_time_us = 0;
   th->major_page_faults = 0;
   th->minor_page_faults = 0;
   th->max_resident_size = 0;
   th->inv_context_switches = 0;
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
   init_papi(th);
 }
@@ -316,7 +316,7 @@ dump_wqueue_counters_single (wstream_df_thread_p th)
 	int i;
 	const char* events[] = WS_PAPI_EVENTS;
 #endif // defined(WS_PAPI_PROFILE)
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
 	printf ("Thread %d: system_time_us = %lld\n",
 		th->worker_id,
 		th->system_time_us);
@@ -332,7 +332,7 @@ dump_wqueue_counters_single (wstream_df_thread_p th)
 	printf ("Thread %d: inv_context_switches = %lld\n",
 		th->worker_id,
 		th->inv_context_switches);
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 	printf ("Thread %d: tasks_created = %lld\n",
 		th->worker_id,
 		th->tasks_created);

libworkstream_df/profiling.h

@@ -49,16 +49,16 @@ init_papi(struct wstream_df_thread* th);
 #define update_papi_timestamp(th, ts) do { } while(0)
 #endif
 
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
 #define PROFILE_RUSAGE_FIELDS \
 	unsigned long long system_time_us; \
 	unsigned long long major_page_faults; \
 	unsigned long long minor_page_faults; \
 	unsigned long long max_resident_size; \
 	unsigned long long inv_context_switches;
-#else // !defined(PROFILE_RUSAGE)
+#else // !PROFILE_RUSAGE
 #define PROFILE_RUSAGE_FIELDS
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
 #if WQUEUE_PROFILE 
 #define WSTREAM_DF_THREAD_WQUEUE_PROFILE_BASIC_FIELDS \
@@ -85,11 +85,11 @@ stop_wqueue_counters (void);
 void
 wqueue_counters_enter_runtime(struct wstream_df_thread* th);
 
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
 void wqueue_counters_profile_rusage(struct wstream_df_thread *th);
-#else // !defined(PROFILE_RUSAGE)
+#else // !PROFILE_RUSAGE
   #define wqueue_counters_profile_rusage(th) do {} while(0)
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
 void
 dump_wqueue_counters (unsigned int num_workers, struct wstream_df_thread** wstream_df_worker_threads);
@@ -127,7 +127,7 @@ dump_global_wqueue_counters ();
 #define set_wqueue_counter_if_zero(ctr, val) do {} while(0)
 #endif
 
-#ifdef MATRIX_PROFILE
+#if MATRIX_PROFILE
 
 extern void *tm_data__;
 #define transfer_matrix ((unsigned long long(*)[wstream_num_workers])tm_data__)
@@ -141,7 +141,7 @@ inline void inc_transfer_matrix_entry(unsigned int consumer,
 void init_transfer_matrix(void);
 void dump_transfer_matrix(unsigned int num_workers);
 
-#else // !defined(MATRIX_PROFILE)
+#else // !MATRIX_PROFILE
 
 #define inc_transfer_matrix_entry(consumer, producer, num_bytes) do {} while(0)
 #define init_transfer_matrix() do {} while(0)

libworkstream_df/reuse.c

@@ -72,7 +72,7 @@ void match_reuse_input_clause_with_output_clause(wstream_df_view_p out_view, wst
   reuse_view_sanity_check(out_view, in_view);
   in_view->reuse_data_view = NULL;
 
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
   out_view->consumer_view = in_view;
 #else
   wstream_df_thread_p cthread = current_thread;
@@ -94,7 +94,7 @@ void match_reuse_output_clause_with_input_clause(wstream_df_view_p out_view, wst
   assert(!is_reuse_view(in_view));
   assert(in_view->refcount == 1);
 
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
   if(!in_view->data) {
     match_reuse_output_clause_with_reuse_input_clause(out_view, in_view);
     return;
@@ -244,7 +244,7 @@ void __built_in_wstream_df_prepare_data_vec(size_t n, void* v)
 
 void __built_in_wstream_df_reuse_update_data(void* v)
 {
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
     const int deferred_alloc_enabled = 1;
 #else
     const int deferred_alloc_enabled = 0;

libworkstream_df/trace.c

@@ -70,13 +70,13 @@ void trace_runtime_counters(struct wstream_df_thread* cthread)
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_SLAB_REFILLS, cthread->slab_cache->slab_refills);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_REUSE_ADDR, cthread->reuse_addr);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_REUSE_COPY, cthread->reuse_copy);
-#ifdef PROFILE_RUSAGE
+#if PROFILE_RUSAGE
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_SYSTEM_TIME_US, cthread->system_time_us);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_MAJOR_PAGE_FAULTS, cthread->major_page_faults);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_MINOR_PAGE_FAULTS, cthread->minor_page_faults);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_MAX_RESIDENT_SIZE, cthread->max_resident_size);
   trace_counter(cthread, RUNTIME_COUNTER_BASE+RUNTIME_COUNTER_INV_CONTEXT_SWITCHES, cthread->inv_context_switches);
-#endif // defined(PROFILE_RUSAGE)
+#endif // PROFILE_RUSAGE
 
   uint64_t steals = 0;
   for(int level = 0; level < MEM_NUM_LEVELS; level++)

libworkstream_df/wstream_df.c

@@ -558,7 +558,7 @@ void __built_in_wstream_df_alloc_view_data(void* v, size_t size)
 
 void __built_in_wstream_df_alloc_view_data_deferred(void* v, size_t size)
 {
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
 	wstream_df_view_p view = v;
 	view->data = NULL;
 #else
@@ -761,7 +761,7 @@ wstream_df_resolve_dependences (void *v, void *s, bool is_read_view_p)
 		  match_reuse_output_clause_with_input_clause(prod_view, view);
 	      }
 
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
 	      /* Data of the consumer view has not been allocated
 	       * yet. If we are the only producer, we further defer
 	       * allocation until the producer gets ready. Otherwise
@@ -828,7 +828,7 @@ wstream_df_resolve_dependences (void *v, void *s, bool is_read_view_p)
 	      match_reuse_output_clause_with_input_clause(view, cons_view);
 	  }
 
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
 	  /* Data of the consumer view has not been allocated
 	   * yet. If we are the only producer, we further defer
 	   * allocation until the producer gets ready. Otherwise
@@ -970,7 +970,7 @@ __attribute__((__optimize__("O1"))) static void worker_thread(void) {
       wqueue_counters_enter_runtime(current_thread);
       inc_wqueue_counter(&cthread->tasks_executed, 1);
     } else {
-#ifndef WS_NO_YIELD_SPIN
+#if WS_NO_YIELD_SPIN
       sched_yield();
 #endif
     }
@@ -1188,15 +1188,15 @@ void pre_main()
 
   init_transfer_matrix();
 
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
   fprintf(stdout, "\n[HWLOC Info] Using %u workers distributed as:\n", wstream_num_workers);
 #endif
   hwloc_obj_t *processor_mapping = NULL;
   if (!distribute_worker_on_topology(wstream_num_workers, &processor_mapping)) {
     wstream_df_error("[hwloc] Warning: could distribute workers on %d CPUs\n",
                      wstream_num_workers);
   }
-#ifdef HWLOC_VERBOSE
+#if HWLOC_VERBOSE
   fprintf(stdout, "\n[HWLOC Info] Worker placement topology:\n");
   print_topology_tree(stdout);
 #endif
@@ -1532,7 +1532,7 @@ broadcast (void *v)
 	  bt->refcount++;
 #endif
       } else {
-#ifdef DEFERRED_ALLOC
+#if DEFERRED_ALLOC
 	  if(!peek_view->data)
 	    __built_in_wstream_df_alloc_view_data(peek_view, peek_view->horizon);
 #endif