[2.26][COMMITTED 1/3] Inline tcache functions

[2.26][COMMITTED 1/3] Inline tcache functions

[Siddhesh Poyarekar]

From: Wilco Dijkstra <wdijkstr@arm.com>

The functions tcache_get and tcache_put show up in profiles as they
are a critical part of the tcache code.  Inline them to give tcache
a 16% performance gain.  Since this improves multi-threaded cases
as well, it helps offset any potential performance loss due to adding
single-threaded fast paths.

	* malloc/malloc.c (tcache_put): Inline.
	(tcache_get): Inline.

(cherry-picked from commit e4dd4ace56880d2f1064cd787e2bdb96ddacc3c4)
---
 ChangeLog       | 5 +++++
 malloc/malloc.c | 4 ++--
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index e17a716..192acbf 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2017-10-17  Wilco Dijkstra  <wdijkstr@arm.com>
+
+	* malloc/malloc.c (tcache_put): Inline.
+	(tcache_get): Inline.
+
 2017-10-13  James Clarke  <jrtc27@jrtc27.com>
 
 	* sysdeps/powerpc/powerpc32/dl-machine.h (elf_machine_rela):
diff --git a/malloc/malloc.c b/malloc/malloc.c
index 01ec157..546579b 100644
--- a/malloc/malloc.c
+++ b/malloc/malloc.c
@@ -2957,7 +2957,7 @@ static __thread tcache_perthread_struct *tcache = NULL;
 
 /* Caller must ensure that we know tc_idx is valid and there's room
    for more chunks.  */
-static void
+static __always_inline void
 tcache_put (mchunkptr chunk, size_t tc_idx)
 {
   tcache_entry *e = (tcache_entry *) chunk2mem (chunk);
@@ -2969,7 +2969,7 @@ tcache_put (mchunkptr chunk, size_t tc_idx)
 
 /* Caller must ensure that we know tc_idx is valid and there's
    available chunks to remove.  */
-static void *
+static __always_inline void *
 tcache_get (size_t tc_idx)
 {
   tcache_entry *e = tcache->entries[tc_idx];
-- 
2.7.5

[2.26][COMMITTED 3/3] Fix build failure on tilepro due to unsupported atomics

[Siddhesh Poyarekar]

From: Wilco Dijkstra <wdijkstr@arm.com>

        * malloc/malloc.c (malloc_state): Use int for have_fastchunks since
        not all targets support atomics on bool.

(cherry-picked from 2c2245b92ccf6344b324d17d8f94ccd3b8c559c6)
---
 ChangeLog       | 5 +++++
 malloc/malloc.c | 3 ++-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/ChangeLog b/ChangeLog
index 67d3503..d67ad03 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2017-10-18  Wilco Dijkstra  <wdijkstr@arm.com>
+
+	* malloc/malloc.c (malloc_state): Use int for have_fastchunks since
+	not all targets support atomics on bool.
+
 2017-10-17  Wilco Dijkstra  <wdijkstr@arm.com>
 
 	* malloc/malloc.c (FASTCHUNKS_BIT): Remove.
diff --git a/malloc/malloc.c b/malloc/malloc.c
index 37e3c44..dd9f699 100644
--- a/malloc/malloc.c
+++ b/malloc/malloc.c
@@ -1689,7 +1689,8 @@ struct malloc_state
   int flags;
 
   /* Set if the fastbin chunks contain recently inserted free blocks.  */
-  bool have_fastchunks;
+  /* Note this is a bool but not all targets support atomics on booleans.  */
+  int have_fastchunks;
 
   /* Fastbins */
   mfastbinptr fastbinsY[NFASTBINS];
-- 
2.7.5

[2.26][COMMITTED 2/3] Use relaxed atomics for malloc have_fastchunks

[Siddhesh Poyarekar]

Currently free typically uses 2 atomic operations per call.  The have_fastchunks
flag indicates whether there are recently freed blocks in the fastbins.  This
is purely an optimization to avoid calling malloc_consolidate too often and
avoiding the overhead of walking all fast bins even if all are empty during a
sequence of allocations.  However using catomic_or to update the flag is
completely unnecessary since it can be changed into a simple boolean and
accessed using relaxed atomics.  There is no change in multi-threaded behaviour
given the flag is already approximate (it may be set when there are no blocks in
any fast bins, or it may be clear when there are free blocks that could be
consolidated).

Performance of malloc/free improves by 27% on a simple benchmark on AArch64
(both single and multithreaded). The number of load/store exclusive instructions
is reduced by 33%. Bench-malloc-thread speeds up by ~3% in all cases.

	* malloc/malloc.c (FASTCHUNKS_BIT): Remove.
	(have_fastchunks): Remove.
	(clear_fastchunks): Remove.
	(set_fastchunks): Remove.
	(malloc_state): Add have_fastchunks.
	(malloc_init_state): Use have_fastchunks.
	(do_check_malloc_state): Remove incorrect invariant checks.
	(_int_malloc): Use have_fastchunks.
	(_int_free): Likewise.
	(malloc_consolidate): Likewise.

(cherry-picked from e956075a5a2044d05ce48b905b10270ed4a63e87)
---
 ChangeLog       | 13 +++++++++++++
 malloc/malloc.c | 52 ++++++++++++++++++++--------------------------------
 2 files changed, 33 insertions(+), 32 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 192acbf..67d3503 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,18 @@
 2017-10-17  Wilco Dijkstra  <wdijkstr@arm.com>
 
+	* malloc/malloc.c (FASTCHUNKS_BIT): Remove.
+	(have_fastchunks): Remove.
+	(clear_fastchunks): Remove.
+	(set_fastchunks): Remove.
+	(malloc_state): Add have_fastchunks.
+	(malloc_init_state): Use have_fastchunks.
+	(do_check_malloc_state): Remove incorrect invariant checks.
+	(_int_malloc): Use have_fastchunks.
+	(_int_free): Likewise.
+	(malloc_consolidate): Likewise.
+
+2017-10-17  Wilco Dijkstra  <wdijkstr@arm.com>
+
 	* malloc/malloc.c (tcache_put): Inline.
 	(tcache_get): Inline.
 
diff --git a/malloc/malloc.c b/malloc/malloc.c
index 546579b..37e3c44 100644
--- a/malloc/malloc.c
+++ b/malloc/malloc.c
@@ -1613,27 +1613,6 @@ typedef struct malloc_chunk *mfastbinptr;
 #define FASTBIN_CONSOLIDATION_THRESHOLD  (65536UL)
 
 /*
-   Since the lowest 2 bits in max_fast don't matter in size comparisons,
-   they are used as flags.
- */
-
-/*
-   FASTCHUNKS_BIT held in max_fast indicates that there are probably
-   some fastbin chunks. It is set true on entering a chunk into any
-   fastbin, and cleared only in malloc_consolidate.
-
-   The truth value is inverted so that have_fastchunks will be true
-   upon startup (since statics are zero-filled), simplifying
-   initialization checks.
- */
-
-#define FASTCHUNKS_BIT        (1U)
-
-#define have_fastchunks(M)     (((M)->flags & FASTCHUNKS_BIT) == 0)
-#define clear_fastchunks(M)    catomic_or (&(M)->flags, FASTCHUNKS_BIT)
-#define set_fastchunks(M)      catomic_and (&(M)->flags, ~FASTCHUNKS_BIT)
-
-/*
    NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous
    regions.  Otherwise, contiguity is exploited in merging together,
    when possible, results from consecutive MORECORE calls.
@@ -1690,6 +1669,17 @@ get_max_fast (void)
    ----------- Internal state representation and initialization -----------
  */
 
+/*
+   have_fastchunks indicates that there are probably some fastbin chunks.
+   It is set true on entering a chunk into any fastbin, and cleared early in
+   malloc_consolidate.  The value is approximate since it may be set when there
+   are no fastbin chunks, or it may be clear even if there are fastbin chunks
+   available.  Given it's sole purpose is to reduce number of redundant calls to
+   malloc_consolidate, it does not affect correctness.  As a result we can safely
+   use relaxed atomic accesses.
+ */
+
+
 struct malloc_state
 {
   /* Serialize access.  */
@@ -1698,6 +1688,9 @@ struct malloc_state
   /* Flags (formerly in max_fast).  */
   int flags;
 
+  /* Set if the fastbin chunks contain recently inserted free blocks.  */
+  bool have_fastchunks;
+
   /* Fastbins */
   mfastbinptr fastbinsY[NFASTBINS];
 
@@ -1841,7 +1834,7 @@ malloc_init_state (mstate av)
   set_noncontiguous (av);
   if (av == &main_arena)
     set_max_fast (DEFAULT_MXFAST);
-  av->flags |= FASTCHUNKS_BIT;
+  atomic_store_relaxed (&av->have_fastchunks, false);
 
   av->top = initial_top (av);
 }
@@ -2206,11 +2199,6 @@ do_check_malloc_state (mstate av)
         }
     }
 
-  if (total != 0)
-    assert (have_fastchunks (av));
-  else if (!have_fastchunks (av))
-    assert (total == 0);
-
   /* check normal bins */
   for (i = 1; i < NBINS; ++i)
     {
@@ -3701,7 +3689,7 @@ _int_malloc (mstate av, size_t bytes)
   else
     {
       idx = largebin_index (nb);
-      if (have_fastchunks (av))
+      if (atomic_load_relaxed (&av->have_fastchunks))
         malloc_consolidate (av);
     }
 
@@ -4116,7 +4104,7 @@ _int_malloc (mstate av, size_t bytes)
 
       /* When we are using atomic ops to free fast chunks we can get
          here for all block sizes.  */
-      else if (have_fastchunks (av))
+      else if (atomic_load_relaxed (&av->have_fastchunks))
         {
           malloc_consolidate (av);
           /* restore original bin index */
@@ -4242,7 +4230,7 @@ _int_free (mstate av, mchunkptr p, int have_lock)
 
     free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
 
-    set_fastchunks(av);
+    atomic_store_relaxed (&av->have_fastchunks, true);
     unsigned int idx = fastbin_index(size);
     fb = &fastbin (av, idx);
 
@@ -4393,7 +4381,7 @@ _int_free (mstate av, mchunkptr p, int have_lock)
     */
 
     if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) {
-      if (have_fastchunks(av))
+      if (atomic_load_relaxed (&av->have_fastchunks))
 	malloc_consolidate(av);
 
       if (av == &main_arena) {
@@ -4464,7 +4452,7 @@ static void malloc_consolidate(mstate av)
   */
 
   if (get_max_fast () != 0) {
-    clear_fastchunks(av);
+    atomic_store_relaxed (&av->have_fastchunks, false);
 
     unsorted_bin = unsorted_chunks(av);
 
-- 
2.7.5