Keep track of when the caches are dirty, and only flush them when
the framebuffer state is set and when the context is flushed.
Signed-off-by: Dave Airlie <airlied@redhat.com>
r600g: modify block to only emit the first few dirty registers. (v2)
This gets me from 2200 to 1978 dwords for a gears frame.
This is due to us having some 32-dwords blocks in the SPI, that we only
modify the first dwords off.
v2: fix dirty reg count from Bas Nieuwenhuizen
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is a first step to decreasing the CPU usage, by decreasing how much
stuff we pass to the GPU and hence to the kernel CS checker.
This adds a check to see if the values we need to write are actually dirty,
and avoids writing if they are. However certain register need to always
be written so we add a new flag to say which ones should be always written
if used. (Note this could probably be done cleaner with a larger refactoring,
since I think the CONST_BUFFER_SIZE_PS/VS and CONST_CACHE_PS/VS might
be better off as a special state).
It also moves the need_bo to be a flags on the register now.
With this, a frame of gears goes from emitting 3k dwords to emitting 2k dwords,
and I'm sure it could get a lot smaller.
v2: fix some evergreen dirty bits.
Original patch from: Bas Nieuwenhuizen, I NIHed nearly the same thing
before seeing his patch on the list, oops.
Reviewed-by: Bas Nieuwenhuizen
Signed-off-by: Dave Airlie <airlied@redhat.com>
The very presence of this extension breaks things.
This should bring us closer to being able to run Unigine Heaven.
The extension will be re-enabled once gl_InstanceID is implemented.
gallium: include u_format.h for util_format_is_supported
Without it gcc complains:
nv50_screen.c: In function ‘nv50_screen_is_format_supported’:
nv50_screen.c:48: warning: implicit declaration of function ‘util_format_is_supported’
and handles it wrongly - util_format_is_supported returns boolean, which is typedef'ed
to uchar, but function without prototype is assumed to return int.
For me nv50_screen_is_format_supported was returning true for float formats without
--enable-texture-float...
nvc0: work around what looks like a code prefetch bug
Sounds very unlikely, but I don't have a better explanation at the
moment.
The GPU throws page faults at the first page after the code buffer
quite frequently on startup, and traces don't show us overflowing.
svga: Rebind framebuffer and tss bindings strictly when necessary.
The earlier change to ensure rendertargets and textures are always
rebound at every command buffer start had the downside of making
successive flushes no longer no-ops, as a command buffer with merely
the rebinding commands were being unnecessarily sent to the vGPU.
This change only re-emits the bindings when necessary, by keeping track
of the need to rebind outside of the dirty state update mechanism.
Fixes issues in e.g. nexuiz (desertfactory) or supertuxkart that
look like lighting bugs.
They're not visible with the software rasterizers because their
notion of linear interpolation seems to be different from that
of nv50/nvc0.
Revert "r300g: handle DISCARD_WHOLE_RESOURCE for buffers"
This reverts commit 437c748bf5.
The commit is wrong for several reasons. One of them is when we grab
a new buffer, we should update all the states it is bound in,
including all parallel contexts. I don't think this is even doable.
The correct solution would be upload data via a temporary buffer and
do resource_copy_region to the original one.
https://bugs.freedesktop.org/show_bug.cgi?id=36088
st/python: Remove bindings, and all its dependencies.
Because:
- bindings are not fully automatic, and they are broken most of the time
- unit tests/samples can be written in C on top of graw
- tracing/retracing is more useful at API levels with stable ABIs such as
GL, producing traces that cover more layers of the driver stack and and
can be used for regression testing
Marek Olšák