bug#21028: Performance regression in revision af1a69f4d17a482c359d98c00ef86fac835b5fac (Apr 2014).

["Clément Pit--Claudel" <clement.pitclaudel@live.com>]

[-- Attachment #1.1: Type: text/plain, Size: 3596 bytes --]

On 03/03/2016 12:08 AM, Clément Pit--Claudel wrote:
> I've reattached both profiles.

Hi,

I've looked further into this. Comparing the profiles, there seems to be two significant differences:

1. ‘deliver_user_signal’ gets called constantly in the slow case, with many calls to ‘x_catch_errors_with_handler’, ‘x_catch_errors_with_handler’, ‘x_had_errors_p’, and ‘x_uncatch_errors’; there are few to no such calls in the fast case.
2. font_list_entities constantly calls xfont_list in the slow case, but not in the fast case:

In the slow case:

    -----------------------------------------------
                    0.04    1.01  554734/554734      font_find_for_lface [14]
    [16]    11.0    0.04    1.01  554734         font_list_entities [16]
                    0.04    0.91  554734/554734      xfont_list [17]
                    0.02    0.00  554734/2778892     Fnreverse [45]
                    0.00    0.02 1109468/1109469     assoc_no_quit [96]
                    0.01    0.00 1109468/1109470     xfont_get_cache [121]
                    0.01    0.00  555016/1289625     Fcons [80]
                    0.00    0.00     141/141         xftfont_list [212]
                    0.00    0.00     141/45490       concat [103]
                    0.00    0.00     141/138863      copy_font_spec [78]
                    0.00    0.00     141/6814        Fvconcat [832]
    -----------------------------------------------

In the fast case:

    -----------------------------------------------
                    0.06    0.41 1533134/1533134     font_find_for_lface [13]
    [25]    14.8    0.06    0.41 1533134         font_list_entities [25]
                    0.02    0.31 3066268/3066269     assoc_no_quit [29]
                    0.07    0.00 1533134/7672938     Fnreverse [28]
                    0.02    0.00 1534750/3266575     Fcons [62]
                    0.00    0.00     347/347         xftfont_list [209]
                    0.00    0.00     461/461         xfont_list [245]
                    0.00    0.00     808/385139      copy_font_spec [50]
                    0.00    0.00     347/56861       concat [111]
                    0.00    0.00 3066268/3066270     xfont_get_cache [605]
                    0.00    0.00     347/5788        Fvconcat [739]
    -----------------------------------------------

The difference there is this line in the slow profile:

    0.04    0.91  554734/554734      xfont_list [17]

(compare to this line:)

    0.00    0.00     461/461         xfont_list [245]

font_list_entities is in src/font.c; it call xfont_list here (old code):

    Lisp_Object copy;

    val = driver_list->driver->list (f, scratch_font_spec); // ← HERE
    if (NILP (val))
      val = zero_vector;
    else
      val = Fvconcat (1, &val);
    copy = copy_font_spec (scratch_font_spec);
    ASET (copy, FONT_TYPE_INDEX, driver_list->driver->type);
    XSETCDR (cache, Fcons (Fcons (copy, val), XCDR (cache)));

Here is the new code:

    val = driver_list->driver->list (f, scratch_font_spec);
    if (!NILP (val))
      {
        Lisp_Object copy = copy_font_spec (scratch_font_spec);

        val = Fvconcat (1, &val);
        ASET (copy, FONT_TYPE_INDEX, driver_list->driver->type);
        XSETCDR (cache, Fcons (Fcons (copy, val), XCDR (cache)));
      }

The commit message didn't explain why it was a good thing to not add an empty vector to the cache.

I'm a bit stuck at this point. I'm not sure how 1 and 2 are connected. Suggestions?

Clément.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]