Merge in r5435 from COMPVBITS. Also added a note to
docs/internals/performance.txt about it.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@5438 a5019735-40e9-0310-863c-91ae7b9d1cf9
diff --git a/coregrind/m_execontext.c b/coregrind/m_execontext.c
index c5b5c93..2b99e92 100644
--- a/coregrind/m_execontext.c
+++ b/coregrind/m_execontext.c
@@ -46,7 +46,8 @@
struct _ExeContext {
struct _ExeContext * next;
- /* Variable-length array. The size is VG_(clo_backtrace_size); at
+ UInt n_ips;
+ /* Variable-length array. The size is 'n_ips'; at
least 1, at most VG_DEEPEST_BACKTRACE. [0] is the current IP,
[1] is its caller, [2] is the caller of [1], etc. */
Addr ips[0];
@@ -126,38 +127,42 @@
/* Print an ExeContext. */
void VG_(pp_ExeContext) ( ExeContext* ec )
{
- VG_(pp_StackTrace)( ec->ips, VG_(clo_backtrace_size) );
+ VG_(pp_StackTrace)( ec->ips, ec->n_ips );
}
/* Compare two ExeContexts, comparing all callers. */
Bool VG_(eq_ExeContext) ( VgRes res, ExeContext* e1, ExeContext* e2 )
{
+ Int i;
+
if (e1 == NULL || e2 == NULL)
return False;
+
+ // Must be at least one address in each trace.
+ tl_assert(e1->n_ips >= 1 && e2->n_ips >= 1);
+
switch (res) {
case Vg_LowRes:
/* Just compare the top two callers. */
ec_cmp2s++;
- if (e1->ips[0] != e2->ips[0]) return False;
-
- if (VG_(clo_backtrace_size) < 2) return True;
- if (e1->ips[1] != e2->ips[1]) return False;
+ for (i = 0; i < 2; i++) {
+ if ( (e1->n_ips <= i) && (e2->n_ips <= i)) return True;
+ if ( (e1->n_ips <= i) && !(e2->n_ips <= i)) return False;
+ if (!(e1->n_ips <= i) && (e2->n_ips <= i)) return False;
+ if (e1->ips[i] != e2->ips[i]) return False;
+ }
return True;
case Vg_MedRes:
/* Just compare the top four callers. */
ec_cmp4s++;
- if (e1->ips[0] != e2->ips[0]) return False;
-
- if (VG_(clo_backtrace_size) < 2) return True;
- if (e1->ips[1] != e2->ips[1]) return False;
-
- if (VG_(clo_backtrace_size) < 3) return True;
- if (e1->ips[2] != e2->ips[2]) return False;
-
- if (VG_(clo_backtrace_size) < 4) return True;
- if (e1->ips[3] != e2->ips[3]) return False;
+ for (i = 0; i < 4; i++) {
+ if ( (e1->n_ips <= i) && (e2->n_ips <= i)) return True;
+ if ( (e1->n_ips <= i) && !(e2->n_ips <= i)) return False;
+ if (!(e1->n_ips <= i) && (e2->n_ips <= i)) return False;
+ if (e1->ips[i] != e2->ips[i]) return False;
+ }
return True;
case Vg_HighRes:
@@ -188,18 +193,20 @@
UWord hash;
ExeContext* new_ec;
ExeContext* list;
+ UInt n_ips;
init_ExeContext_storage();
- vg_assert(VG_(clo_backtrace_size) >= 1
- && VG_(clo_backtrace_size) <= VG_DEEPEST_BACKTRACE);
+ vg_assert(VG_(clo_backtrace_size) >= 1 &&
+ VG_(clo_backtrace_size) <= VG_DEEPEST_BACKTRACE);
- VG_(get_StackTrace)( tid, ips, VG_(clo_backtrace_size) );
+ n_ips = VG_(get_StackTrace)( tid, ips, VG_(clo_backtrace_size) );
+ tl_assert(n_ips >= 1);
/* Now figure out if we've seen this one before. First hash it so
as to determine the list number. */
hash = 0;
- for (i = 0; i < VG_(clo_backtrace_size); i++) {
+ for (i = 0; i < n_ips; i++) {
hash ^= ips[i];
hash = (hash << 29) | (hash >> 3);
}
@@ -215,7 +222,7 @@
if (list == NULL) break;
ec_searchcmps++;
same = True;
- for (i = 0; i < VG_(clo_backtrace_size); i++) {
+ for (i = 0; i < n_ips; i++) {
if (list->ips[i] != ips[i]) {
same = False;
break;
@@ -234,13 +241,14 @@
ec_totstored++;
new_ec = VG_(arena_malloc)( VG_AR_EXECTXT,
- sizeof(struct _ExeContext *)
- + VG_(clo_backtrace_size) * sizeof(Addr) );
+ sizeof(struct _ExeContext)
+ + n_ips * sizeof(Addr) );
- for (i = 0; i < VG_(clo_backtrace_size); i++)
+ for (i = 0; i < n_ips; i++)
new_ec->ips[i] = ips[i];
- new_ec->next = ec_list[hash];
+ new_ec->n_ips = n_ips;
+ new_ec->next = ec_list[hash];
ec_list[hash] = new_ec;
return new_ec;