ensure that get_timing_subsystem and DISABLE_SUBSYSTEM agree on u…

…sing sorted indices

previously, `get_timing_subsystem` used sorted indices while `DISABLE_SUBSYSTEM` used non sorted which lead to the wrong zones getting disabled.

src/timing.c

@@ -94,6 +94,16 @@ static const int indirect_strcmp(const void *a, const void *b) {
     return strcmp(*(const char **)a, *(const char **)b);
 }
 
+static inline int get_timing_subsystem(const char *subsystem) {
+    const char **match = (const char **)bsearch(
+        &subsystem, jl_timing_subsystems, JL_TIMING_SUBSYSTEM_LAST,
+        sizeof(const char *), indirect_strcmp
+    );
+    if (!match)
+        return JL_TIMING_SUBSYSTEM_LAST;
+    return (int)(match - &jl_timing_subsystems[0]);
+}
+
 void jl_init_timing(void)
 {
     t0 = cycleclock();
@@ -154,7 +164,11 @@ void jl_init_timing(void)
  * files, so we disable them by default.
  **/
 #ifdef DISABLE_FREQUENT_EVENTS
-#define DISABLE_SUBSYSTEM(subsystem) jl_atomic_fetch_or_relaxed(jl_timing_disable_mask + (JL_TIMING_##subsystem / (sizeof(uint64_t) * CHAR_BIT)), 1 << (JL_TIMING_##subsystem % (sizeof(uint64_t) * CHAR_BIT)))
+#define DISABLE_SUBSYSTEM(subsystem) \
+    do { \
+        int sorted_subsystem = get_timing_subsystem(#subsystem);             \
+        jl_atomic_fetch_or_relaxed(jl_timing_disable_mask + (sorted_subsystem / (sizeof(uint64_t) * CHAR_BIT)), (uint64_t)1 << (sorted_subsystem % (sizeof(uint64_t) * CHAR_BIT))); \
+    } while (0)
     DISABLE_SUBSYSTEM(ROOT);
     DISABLE_SUBSYSTEM(TYPE_CACHE_LOOKUP);
     DISABLE_SUBSYSTEM(METHOD_MATCH);
@@ -175,17 +189,6 @@ void jl_destroy_timing(void)
     }
 }
 
-static const int get_timing_subsystem(const char *subsystem) {
-    const char **match = (const char **)bsearch(
-        &subsystem, jl_timing_subsystems, JL_TIMING_SUBSYSTEM_LAST,
-        sizeof(const char *), indirect_strcmp
-    );
-    if (!match)
-        return JL_TIMING_SUBSYSTEM_LAST;
-
-    return (int)(match - &jl_timing_subsystems[0]);
-}
-
 #ifdef USE_TIMING_COUNTS
 
 // This function is analogous to __itt_event_create but for the counts backend

src/timing.h

@@ -317,7 +317,7 @@ struct _jl_timing_block_t { // typedef in julia.h
 };
 
 STATIC_INLINE int _jl_timing_enabled(int subsystem) JL_NOTSAFEPOINT {
-    return (jl_atomic_load_relaxed(jl_timing_disable_mask + subsystem / (sizeof(uint64_t) * CHAR_BIT)) & (1 << (subsystem % (sizeof(uint64_t) * CHAR_BIT)))) == 0;
+    return (jl_atomic_load_relaxed(jl_timing_disable_mask + subsystem / (sizeof(uint64_t) * CHAR_BIT)) & ((uint64_t)1 << (subsystem % (sizeof(uint64_t) * CHAR_BIT)))) == 0;
 }
 
 typedef struct _jl_timing_suspend_t {