Restructure setters for profiling info

Explicitly define three setters:

- `prof_tctx_reset()`: set `prof_tctx` to `1U`, if we don't know in
advance whether the allocation is large or not;
- `prof_tctx_reset_sampled()`: set `prof_tctx` to `1U`, if we already
know in advance that the allocation is large;
- `prof_info_set()`: set a real `prof_tctx`, and also set other
profiling info e.g. the allocation time.

Code structure wise, the prof level is kept as a thin wrapper, the
large level only provides low level setter APIs, and the arena level
carries out the main logic.

include/jemalloc/internal/arena_inlines_b.h

@@ -68,8 +68,7 @@ arena_prof_info_get(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx,
 }
 
 JEMALLOC_ALWAYS_INLINE void
-arena_prof_tctx_set(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx,
+arena_prof_tctx_reset(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx) {
-    prof_tctx_t *tctx) {
 	cassert(config_prof);
 	assert(ptr != NULL);
 
@@ -77,17 +76,17 @@ arena_prof_tctx_set(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx,
 	if (alloc_ctx == NULL) {
 		extent_t *extent = iealloc(tsd_tsdn(tsd), ptr);
 		if (unlikely(!extent_slab_get(extent))) {
-			large_prof_tctx_set(extent, tctx);
+			large_prof_tctx_reset(extent);
 		}
 	} else {
 		if (unlikely(!alloc_ctx->slab)) {
-			large_prof_tctx_set(iealloc(tsd_tsdn(tsd), ptr), tctx);
+			large_prof_tctx_reset(iealloc(tsd_tsdn(tsd), ptr));
 		}
 	}
 }
 
-static inline void
+JEMALLOC_ALWAYS_INLINE void
-arena_prof_tctx_reset(tsd_t *tsd, const void *ptr, prof_tctx_t *tctx) {
+arena_prof_tctx_reset_sampled(tsd_t *tsd, const void *ptr) {
 	cassert(config_prof);
 	assert(ptr != NULL);
 
@@ -98,13 +97,13 @@ arena_prof_tctx_reset(tsd_t *tsd, const void *ptr, prof_tctx_t *tctx) {
 }
 
 JEMALLOC_ALWAYS_INLINE void
-arena_prof_alloc_time_set(tsd_t *tsd, const void *ptr, nstime_t *t) {
+arena_prof_info_set(tsd_t *tsd, const void *ptr, prof_tctx_t *tctx) {
 	cassert(config_prof);
 	assert(ptr != NULL);
 
 	extent_t *extent = iealloc(tsd_tsdn(tsd), ptr);
 	assert(!extent_slab_get(extent));
-	large_prof_alloc_time_set(extent, t);
+	large_prof_info_set(extent, tctx);
 }
 
 JEMALLOC_ALWAYS_INLINE void

include/jemalloc/internal/large_externs.h

@@ -23,8 +23,7 @@ void large_dalloc_finish(tsdn_t *tsdn, extent_t *extent);
 void large_dalloc(tsdn_t *tsdn, extent_t *extent);
 size_t large_salloc(tsdn_t *tsdn, const extent_t *extent);
 void large_prof_info_get(const extent_t *extent, prof_info_t *prof_info);
-void large_prof_tctx_set(extent_t *extent, prof_tctx_t *tctx);
 void large_prof_tctx_reset(extent_t *extent);
-void large_prof_alloc_time_set(extent_t *extent, nstime_t *time);
+void large_prof_info_set(extent_t *extent, prof_tctx_t *tctx);
 
 #endif /* JEMALLOC_INTERNAL_LARGE_EXTERNS_H */

include/jemalloc/internal/prof_inlines_b.h

@@ -50,28 +50,28 @@ prof_info_get(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx,
 }
 
 JEMALLOC_ALWAYS_INLINE void
-prof_tctx_set(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx,
+prof_tctx_reset(tsd_t *tsd, const void *ptr, alloc_ctx_t *alloc_ctx) {
-    prof_tctx_t *tctx) {
 	cassert(config_prof);
 	assert(ptr != NULL);
 
-	arena_prof_tctx_set(tsd, ptr, alloc_ctx, tctx);
+	arena_prof_tctx_reset(tsd, ptr, alloc_ctx);
 }
 
 JEMALLOC_ALWAYS_INLINE void
-prof_tctx_reset(tsd_t *tsd, const void *ptr, prof_tctx_t *tctx) {
+prof_tctx_reset_sampled(tsd_t *tsd, const void *ptr) {
 	cassert(config_prof);
 	assert(ptr != NULL);
 
-	arena_prof_tctx_reset(tsd, ptr, tctx);
+	arena_prof_tctx_reset_sampled(tsd, ptr);
 }
 
 JEMALLOC_ALWAYS_INLINE void
-prof_alloc_time_set(tsd_t *tsd, const void *ptr, nstime_t *t) {
+prof_info_set(tsd_t *tsd, const void *ptr, prof_tctx_t *tctx) {
 	cassert(config_prof);
 	assert(ptr != NULL);
+	assert((uintptr_t)tctx > (uintptr_t)1U);
 
-	arena_prof_alloc_time_set(tsd, ptr, t);
+	arena_prof_info_set(tsd, ptr, tctx);
 }
 
 JEMALLOC_ALWAYS_INLINE bool
@@ -125,8 +125,7 @@ prof_malloc(tsd_t *tsd, const void *ptr, size_t usize, alloc_ctx_t *alloc_ctx,
 	if (unlikely((uintptr_t)tctx > (uintptr_t)1U)) {
 		prof_malloc_sample_object(tsd, ptr, usize, tctx);
 	} else {
-		prof_tctx_set(tsd, ptr, alloc_ctx,
+		prof_tctx_reset(tsd, ptr, alloc_ctx);
-		    (prof_tctx_t *)(uintptr_t)1U);
 	}
 }
 
@@ -161,15 +160,15 @@ prof_realloc(tsd_t *tsd, const void *ptr, size_t usize, prof_tctx_t *tctx,
 	if (unlikely(sampled)) {
 		prof_malloc_sample_object(tsd, ptr, usize, tctx);
 	} else if (moved) {
-		prof_tctx_set(tsd, ptr, NULL, (prof_tctx_t *)(uintptr_t)1U);
+		prof_tctx_reset(tsd, ptr, NULL);
 	} else if (unlikely(old_sampled)) {
 		/*
-		 * prof_tctx_set() would work for the !moved case as well, but
+		 * prof_tctx_reset() would work for the !moved case as well,
-		 * prof_tctx_reset() is slightly cheaper, and the proper thing
+		 * but prof_tctx_reset_sampled() is slightly cheaper, and the
-		 * to do here in the presence of explicit knowledge re: moved
+		 * proper thing to do here in the presence of explicit
-		 * state.
+		 * knowledge re: moved state.
 		 */
-		prof_tctx_reset(tsd, ptr, tctx);
+		prof_tctx_reset_sampled(tsd, ptr);
 	} else {
 		prof_info_t prof_info;
 		prof_info_get(tsd, ptr, NULL, &prof_info);

src/large.c

@@ -372,7 +372,7 @@ large_prof_info_get(const extent_t *extent, prof_info_t *prof_info) {
 	extent_prof_info_get(extent, prof_info);
 }
 
-void
+static void
 large_prof_tctx_set(extent_t *extent, prof_tctx_t *tctx) {
 	extent_prof_tctx_set(extent, tctx);
 }
@@ -383,6 +383,9 @@ large_prof_tctx_reset(extent_t *extent) {
 }
 
 void
-large_prof_alloc_time_set(extent_t *extent, nstime_t *t) {
+large_prof_info_set(extent_t *extent, prof_tctx_t *tctx) {
-	extent_prof_alloc_time_set(extent, t);
+	large_prof_tctx_set(extent, tctx);
+	nstime_t t;
+	nstime_init_update(&t);
+	extent_prof_alloc_time_set(extent, &t);
 }

src/prof.c

@@ -162,13 +162,7 @@ prof_alloc_rollback(tsd_t *tsd, prof_tctx_t *tctx, bool updated) {
 void
 prof_malloc_sample_object(tsd_t *tsd, const void *ptr, size_t usize,
     prof_tctx_t *tctx) {
-	prof_tctx_set(tsd, ptr, NULL, tctx);
+	prof_info_set(tsd, ptr, tctx);
-
-	/* Get the current time and set this in the extent_t. We'll read this
-	 * when free() is called. */
-	nstime_t t;
-	nstime_init_update(&t);
-	prof_alloc_time_set(tsd, ptr, &t);
 
 	malloc_mutex_lock(tsd_tsdn(tsd), tctx->tdata->lock);
 	tctx->cnts.curobjs++;