Re: WIP: Make timestamptz_out less slow.

Hi,

I recently once more noticed that timestamptz_out is really, really
slow. To quantify that, I created a bunch of similar sized tables:

CREATE TABLE tbl_timestamp AS SELECT NOW() FROM generate_series(1, 100000) a, generate_series(1, 100) b;
CREATE TABLE tbl_int8 AS SELECT 1::bigint FROM generate_series(1,  100000) a, generate_series(1, 100) b;
CREATE TABLE tbl_bytea AS SELECT '   '::bytea FROM generate_series(1, 100000) a, generate_series(1, 100) b;

These all end up being 346MB large.

COPY tbl_bytea TO '/dev/null';
Time: 1173.484 ms
COPY tbl_int8 TO '/dev/null';
Time: 1030.756 ms
COPY tbl_timestamp TO '/dev/null';
Time: 6598.030

(all best of three)

Yes, timestamp outputs more data as a whole, but being 5 times as slow
is still pretty darn insane. To make sure that's not the cause, here's
another table:

CREATE TABLE tbl_timestamptext AS SELECT NOW()::text FROM generate_series(1, 100000) a, generate_series(1, 100) b;
COPY tbl_timestamptext TO '/dev/null';
Time: 1449.554 ms

So it's really just the timestamp code.


Profiling it shows:
  Overhead  Command         Shared Object     Symbol
  -   38.33%  postgres_stock  libc-2.19.so      [.] vfprintf
     - 97.92% vfprintf
          _IO_vsprintf
        - sprintf
           + 70.25% EncodeDateTime
           + 29.75% AppendSeconds.constprop.10
     + 1.11% _IO_default_xsputn
  -    8.22%  postgres_stock  libc-2.19.so      [.] _IO_default_xsputn
     - 99.43% _IO_default_xsputn
        - 90.09% vfprintf
             _IO_vsprintf
           - sprintf
              + 74.15% EncodeDateTime
              + 25.85% AppendSeconds.constprop.10
        + 9.72% _IO_padn
     + 0.57% vfprintf
  +   7.76%  postgres_stock  postgres_stock    [.] CopyOneRowTo

So nearly all the time is spent somewhere inside the sprintf calls. Not
nice.

The only thing I could come up to make the sprintfs cheaper was to
combine them into one and remove some of the width specifiers that
aren't needed. That doesn't buy us very much.

I then proceeded to replace the sprintf call with hand-rolled
conversions. And wow, the benefit is far bigger than I'd assumed:
postgres[7236][1]=# COPY tbl_timestamp TO '/dev/null';
Time: 2430.521 ms

So, by hand-rolling the ISO conversion in EncodeDateTime() we got a
~250% performance improvement. I'd say that's worthwhile.

The attached patch shows what I did. While there's some polishing
possible, as a whole, it's pretty ugly. But I think timestamp data is so
common that it's worth the effort.

Does anybody have a fundamentally nicer idea than the attached to
improvide this?