I've tried this git repo:
http://cgit.freedesktop.org/~andrem/mesa/log/?h=r600-blit

Tested performance:

1) Bloom: off && mesa's master:
840 frames 18.5 seconds 45.5 fps 5.0/22.0/225.0/9.7 ms

2) Bloom: on && mesa's master:
untestable

3) Bloom: off && ~andrem's r600-blit
840 frames 18.9 seconds 44.3 fps 6.0/22.6/293.0/10.0 ms

4) Bloom: on && ~andrem's r600-blit
840 frames 31.4 seconds 26.8 fps 24.0/37.4/59.0/6.3 ms

So that patches don't cause performance regression and significantly improve "Bloom: On"! :)

I saw some corruptions (maybe related to:
format for copy MESA_FORMAT_XRGB8888
format for copy MESA_FORMAT_RGBA8888_REV
format for copy MESA_FORMAT_XRGB8888
format for copy MESA_FORMAT_RGBA8888_REV
?) will investigate.

Ignore my corruption suspicion. It's not blit related (see bug #26011).

With "r600: blit - add formats, cleanup, formatting" I don't see that format messages anymore :)

Before "r600: blit - add formats, cleanup, formatting"
840 frames 32.3 seconds 26.0 fps 24.0/38.4/243.0/11.5 ms
840 frames 31.7 seconds 26.5 fps 23.0/37.8/63.0/6.3 ms
840 frames 32.0 seconds 26.3 fps 24.0/38.0/241.0/10.4 ms

After "r600: blit - add formats, cleanup, formatting"
840 frames 27.2 seconds 30.8 fps 19.0/32.4/86.0/6.0 ms
840 frames 28.0 seconds 29.9 fps 20.0/33.4/231.0/13.1 ms
840 frames 27.5 seconds 30.5 fps 20.0/32.8/254.0/10.2 ms

But this may be improvement due to less printing only, I do not know.

should be fixed in mesa master with r600 blit support.

Thanks Andre, thanks Alex.