This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH v3 3/8] btrace: Use binary search to find instruction.


Tim Wiederhake writes:
 > Currently, btrace_find_insn_by_number will iterate over all function call
> segments to find the one that contains the needed instruction. This linear
 > search is too slow for the upcoming Python bindings that will use this
> function to access instructions. This patch introduces a vector in struct > btrace_thread_info that holds pointers to all recorded function segments and
 > allows to use binary search.
 >
> The proper solution is to turn the underlying tree into a vector of objects
 > and use indices for access.  This requires more work.  A patch set is
 > currently being worked on and will be published later.
 >
 > 2016-11-21  Tim Wiederhake  <tim.wiederhake@intel.com>
 >
 > gdb/ChangeLog:
 > 	* btrace.c (btrace_fetch): Copy function call segments pointer
 > 	into a vector.
 > 	(btrace_clear): Clear the vector.
 > 	(btrace_find_insn_by_number): Use binary search to find the correct
 > 	function call segment.
 > 	* btrace.h (brace_fun_p): New typedef.
 > 	(struct btrace_thread_info) <functions>: New field.
 >
 >
 > ---
 >  gdb/btrace.c | 39 +++++++++++++++++++++++++++++++++------
 >  gdb/btrace.h |  7 +++++++
 >  2 files changed, 40 insertions(+), 6 deletions(-)
 >
 > diff --git a/gdb/btrace.c b/gdb/btrace.c
 > index ac7d46e..c878407 100644
 > --- a/gdb/btrace.c
 > +++ b/gdb/btrace.c
 > @@ -1810,6 +1810,8 @@ btrace_fetch (struct thread_info *tp)
 >    /* Compute the trace, provided we have any.  */
 >    if (!btrace_data_empty (&btrace))
 >      {
 > +      struct btrace_function *bfun;
 > +
 >        /* Store the raw trace data.  The stored data will be cleared in
 >  	 btrace_clear, so we always append the new trace.  */
 >        btrace_data_append (&btinfo->data, &btrace);
 > @@ -1817,6 +1819,11 @@ btrace_fetch (struct thread_info *tp)
 >
 >        btrace_clear_history (btinfo);
 >        btrace_compute_ftrace (tp, &btrace);
 > +
 > +      VEC_truncate (btrace_fun_p, btinfo->functions, 0);
 > +
 > +      for (bfun = btinfo->begin; bfun != NULL; bfun = bfun->flow.next)
 > +	VEC_safe_push (btrace_fun_p, btinfo->functions, bfun);
 >      }
 >
 >    do_cleanups (cleanup);
 > @@ -1839,6 +1846,8 @@ btrace_clear (struct thread_info *tp)
 >
 >    btinfo = &tp->btrace;
 >
 > +  VEC_free (btrace_fun_p, btinfo->functions);
 > +
 >    it = btinfo->begin;
 >    while (it != NULL)
 >      {
> @@ -2429,20 +2438,38 @@ btrace_find_insn_by_number (struct btrace_insn_iterator *it,
 >  			    unsigned int number)
 >  {
 >    const struct btrace_function *bfun;
 > +  unsigned int upper, lower, average;
 >
 > -  for (bfun = btinfo->end; bfun != NULL; bfun = bfun->flow.prev)
 > -    if (bfun->insn_offset <= number)
 > -      break;
 > +  if (VEC_empty (btrace_fun_p, btinfo->functions))
 > +      return 0;
 >
 > -  if (bfun == NULL)
 > +  lower = 0;
 > +  bfun = VEC_index (btrace_fun_p, btinfo->functions, lower);
 > +  if (number < bfun->insn_offset)
 >      return 0;
 >
 > -  if (bfun->insn_offset + ftrace_call_num_insn (bfun) <= number)
 > +  upper = VEC_length (btrace_fun_p, btinfo->functions) - 1;
 > +  bfun = VEC_index (btrace_fun_p, btinfo->functions, upper);
 > +  if (number >= bfun->insn_offset + ftrace_call_num_insn (bfun))
 >      return 0;
 >
 > +  average = lower + (upper - lower) / 2;
 > +  bfun = VEC_index (btrace_fun_p, btinfo->functions, average);
 > +
 > +  while ((number >= bfun->insn_offset + ftrace_call_num_insn (bfun))
 > +      || (number < bfun->insn_offset))
 > +    {
 > +      if (number < bfun->insn_offset)
 > +	upper = average - 1;
 > +      else
 > +	lower = average + 1;
 > +
 > +      average = lower + (upper - lower) / 2;
 > +      bfun = VEC_index (btrace_fun_p, btinfo->functions, average);
 > +    }

It's a bit hard to reason about the correctness of this loop
for all edge conditions, but maybe it's just insufficient
familiarity with bfun. Ok to leave as is, just thinking
out loud.

 > +
 >    it->function = bfun;
 >    it->index = number - bfun->insn_offset;
 > -
 >    return 1;
 >  }
 >
 > diff --git a/gdb/btrace.h b/gdb/btrace.h
 > index d0497b4..6de2307 100644
 > --- a/gdb/btrace.h
 > +++ b/gdb/btrace.h
 > @@ -187,6 +187,9 @@ struct btrace_function
 >    btrace_function_flags flags;
 >  };
 >
 > +typedef struct btrace_function *btrace_fun_p;
 > +DEF_VEC_P (btrace_fun_p);
 > +
 >  /* A branch trace instruction iterator.  */
 >  struct btrace_insn_iterator
 >  {
 > @@ -337,6 +340,10 @@ struct btrace_thread_info
 >    struct btrace_function *begin;
 >    struct btrace_function *end;
 >
> + /* Vector of pointer to decoded function segments. These are in execution > + order with the first element == BEGIN and the last element == END. */
 > +  VEC (btrace_fun_p) *functions;
 > +
 >    /* The function level offset.  When added to each function's LEVEL,
 >       this normalizes the function levels such that the smallest level
 >       becomes zero.  */
 > --
 > 2.7.4
 >

LGTM otherwise.
[with same proviso regarding others comments]


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]