xf86-video-freedreno: Changes to 'master'

[Rob Clark]

New branch 'master' available with the following commits:
commit 29f1a057b48089ddbea4aa022dc04fca5467ca0a
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 22:19:50 2013 -0500

    dri2: allocate DRI2 buffers in SMI
    
    Attempt to use the fast stacked-memory for render target buffers.

commit b3b7feca46a6e8de93e6bf7b0b7bd5fce2fd247f
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 20:59:47 2013 -0500

    accel: reduce cmdstream buffer size
    
    The size field in VGV3_NEXTCMD is only 12 bits, so we can submit
    cmdstream packets with payload larger than this.  Reduce the size
    of the cmdstream buffer to compensate.
    
    Also, the smaller size turns up some bugs in handling of ringbuffer
    switch (we'd switch in BEGIN_RING() -> FIRE_RING(), but the exa fxns
    grab the ringbuffer ptr before that), so fix that as well.

commit 056933c87c9122dab36e076e2e844920f311848b
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 21:12:33 2013 -0500

    exa: add support for composite w/ repeat

commit dc8f8cff38a9f7b5fac615bfb5a8150860920f9a
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 21:02:47 2013 -0500

    exa: use CreatePixmap2
    
    This will be handy later if we want DRI to request creation of render
    targets in SMI.

commit 45bdc281e05781989d90bf6fe00c0d3450257f7b
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 20:57:13 2013 -0500

    add libdrm debugging/traces
    
    Normally this gets registered in xserver DRI1 code, but we don't support
    DRI1.

commit 4ca701ecf72e79203db95668531e4e99bb6d4325
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Mar 6 20:50:59 2013 -0500

    update registers header

commit 026c9916217538cb734fe27ef54df366c46a22cb
Author: Rob Clark <robdclark at gmail.com>
Date:   Mon Feb 25 16:31:41 2013 -0500

    exa: composite ops w/ mask and other cleanups

commit 8f3f558a46e3495fa3df2ac68ecd181f59f2b8f0
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Dec 16 10:43:54 2012 -0600

    update for compat-api

commit dc5221a65749ba645ae2f38c7a48143904fe093e
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 21 14:32:19 2012 -0500

    add dri2

commit b93fff5f44d7d4f05f44200f08af97cea33e1eaf
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 21 12:20:07 2012 -0500

    update fd_ringbuffer_emit_reloc()
    
    for new arg

commit c7a22a31986f257bdf3c6fd23b39385884fff25a
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Oct 20 17:19:54 2012 -0500

    add batching
    
    This increases performance significantly.  There is still room for
    performance (avoiding sw fallbacks, and possibly we can optimize
    the cmdstream emission), but it should be usable now.

commit 79397b5f328439e5abae746a6e917a18bd1cbf43
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Oct 13 11:32:59 2012 -0500

    convert to use libdrm_freedreno

commit 24aead9e723ba88e684457678189b223d8758312
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 11:50:51 2012 -0500

    exa: begin cleaning up hard-coded constants
    
    The basic cmdstream format seems to be:
    
       (regaddr << 24) | regval
    
    with regaddr of 0x7c triggering the next 'count' dwords as a series
    of register values (which allows writing values >24bit).

commit 004080796707c3cd2ac2864e04e4262515f43cff
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 10:38:34 2012 -0500

    exa: move scissor x/y into out_dstpix()
    
    For copy/composite, we were setting this to be the entire pixmap.. but
    this should be ok for fill as well, so simplify the code by moving it
    into out_dstpix().

commit 313b8ac299fd9a191b02d31a446536a5e44c8101
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 08:56:11 2012 -0500

    driver: fix issue w/ lightdm
    
    We might not be master yet, in which case drmSetInterfaceVersion()
    would fail.  Don't be a crybaby about it.

commit f77054c9b20b958dd26a33ceb71538d2f06d049f
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 08:55:32 2012 -0500

    driver: improve panning/backlight hack
    
    So we can run X11 twice without rebooting.

commit 4d05b6f73a971e558b9d129cffe9d7f6612be705
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 08:32:03 2012 -0500

    accel: turn off dword logging by default
    
    It is a bit too much logging.

commit 7cb364856cbeca22931f1b7d9d0781200440c9b8
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Oct 7 08:30:31 2012 -0500

    driver: hack to make sure backlight gets enabled
    
    The msm fbdev driver is made of fail.  It forgets to enable the
    backlight.  Panning the display is enough to kick it so that we
    can actually see what is on the screen.

commit c174fc90ad4699660fbeac0048cc2d4560e6366e
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Apr 7 15:30:37 2012 -0500

    exa: cleanups
    
    Now that solid/copy/composite are basically working, and since there
    seems to be some commonality in how src or dst surface info is emitted,
    refactor this into some helper fxns.  And other such cleanup.

commit f01639d0feabb88078419562a66fb67f4540be8c
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Apr 7 13:29:44 2012 -0500

    exa: composite
    
    Seems to work! \o/
    
    sans-mask.. still need to figure out how to handle that..

commit d10a61e1d8ba4fff577a9c0ced199b2f85b819a7
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Apr 7 10:15:54 2012 -0500

    exa: more work on composite
    
    Getting closer.. still need to figure out how to deal w/ porter-duff op
    encoding, which seems to also be a function of src and dst (but not mask?)
    formats.
    
    Also, need to figure out how to deal with maskX/maskY.. libC2D2 does not
    give a way to specify different src and mask rect, so not quite sure if
    this is a hw limit.. it might be worth experimenting w/ cmdstream to see
    if we can guess how to handle this.  Otherwise, we might be able to fake
    it via ptr arithmatic on the gpuaddr.

commit e8cc64343cef9f475ebaccb891f62f717b47eda7
Author: Rob Clark <robdclark at gmail.com>
Date:   Fri Apr 6 21:29:34 2012 -0500

    exa: begin adding composite
    
    Still a ways from being functional..

commit d4f3db8f1276423c07cc763f8e1d0ce3801afa4a
Author: Rob Clark <robdclark at gmail.com>
Date:   Wed Apr 4 20:13:35 2012 -0500

    exa: add solid copy
    
    Again limit to 32bpp.. that covers the majority of copy blits and we
    can go back later and add other formats.

commit d066cb3a2cd4fbfb440aa4fba6cd3355be36c729
Author: Rob Clark <robdclark at gmail.com>
Date:   Tue Apr 3 19:19:50 2012 -0500

    exa: Add solid fill
    
    For now just limiting to 32bpp.. probably the color format or bpp is
    encoded somewhere, which I need to figure out.

commit e53f8f9a10946632221317f4036611cdfb415bd6
Author: Rob Clark <robdclark at gmail.com>
Date:   Sun Apr 1 22:11:47 2012 -0500

    accel: add debug support to log dwords

commit bafb8113f4f81d774665274b025133e1f551c378
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Mar 31 18:20:00 2012 -0500

    accel: Add ringbuffer and initial context setup

commit 3c130bf17f802da8d5a6bda8351a401f7e9c2cd9
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Mar 31 12:26:20 2012 -0500

    Fix handling for framebuffer bo

commit 085d5e28c8febf48bfa0c4a6471b69e929780ed9
Author: Rob Clark <robdclark at gmail.com>
Date:   Sat Mar 31 12:30:18 2012 -0500

    Initial commit
    
    based on a stripped down xf86-video-msm.. the goal is to get something
    with just EXA and remove the unneeded bits (like xv and dri2) for now.
    They can be re-added later with removal of dependency on c2d2 if needed.