Don't count a leading 1 as a signficant digit in the ratio.

What used to look like this:
            desk_pokemonwiki.skp    9.38ms -> 9.76ms    1x
                 tabl_pravda.skp     237us ->  241us    1x
          desk_css3gradients.skp     249us ->  254us    1x
            ....
               desk_fontwipe.skp    39.6us -> 38.7us    0.98x
                   tabl_digg.skp     922us ->  893us    0.97x
                  tabl_gmail.skp    20.7us ->   20us    0.96x

Now will print more like this:

            desk_pokemonwiki.skp    9.38ms -> 9.76ms    1.04x
                 tabl_pravda.skp     237us ->  241us    1.02x
          desk_css3gradients.skp     249us ->  254us    1.02x
            ....
               desk_fontwipe.skp    39.6us -> 38.7us    0.98x
                   tabl_digg.skp     922us ->  893us    0.97x
                  tabl_gmail.skp    20.7us ->   20us    0.96x

BUG=skia:

Review URL: https://codereview.chromium.org/756643004

1 files changed, 2 insertions, 1 deletions

diff --git a/bin/compare b/bin/compare
index fe489abadf..e911f4a6c3 100755
--- a/bin/compare
+++ b/bin/compare
@@ -36,4 +36,5 @@ maxlen = max(map(len, common))
 bonferroni = SIGNIFICANCE_THRESHOLD / len(ps)  # Adjust for the fact we've run multiple tests.
 for ratio, p, key, am, bm in ps:
     if p < bonferroni:
-        print '%*s\t%6s -> %6s\t%.2gx' % (maxlen, key, humanize(am), humanize(bm), ratio)
+        str_ratio = ('%.2gx' if ratio < 1 else '%.3gx') % ratio
+        print '%*s\t%6s -> %6s\t%s' % (maxlen, key, humanize(am), humanize(bm), str_ratio)