729f1ec786
sna/gen2: Prefer to use the BLT
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-18 10:29:33 +01:00
adef3a56cd
sna/gen3: Prefer to use the BLT where possible for composite ops
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-18 10:29:33 +01:00
f40e6ab50b
sna/gen3: Prefer to use the BLT to clear the scratch glyph pixmaps
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-18 10:29:33 +01:00
941e3504e0
sna: Some more debug output for request/bo retirement
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 21:25:17 +01:00
1d82fe3e52
sna: Only retire the flush handler if nothing was submitted within the period
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 21:25:16 +01:00
7758e33340
sna/gen3: Use immediates for black/white solid sources
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 21:25:16 +01:00
29ca1a3922
sna/gen3: More removal of memset(0)
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 17:59:53 +01:00
88bc108788
sna/gen5: Rearrange the BLT->RENDER workaround
...
So that we can simply use the pending DRAWRECT as the non-pipelined
flush required following use of a BLT command.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 16:44:59 +01:00
18aed47ba3
sna: Fast path common colour conversions
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 16:41:14 +01:00
a3466c8b69
sna/accel: Implement a simpler path for CopyArea between the same pixmaps
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 16:10:31 +01:00
8afa05191a
sna: Improve debug output for mi/fb fallbacks
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 16:08:17 +01:00
2b118658d5
sna/gen3: Remove memset(0) from fill_boxes
...
And initialise only the state used for the operation.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 16:08:17 +01:00
f1bb4ebfd8
sna/damage: Clear damage->all when subtracting
...
I missed clearing the all-damaged flag on the subtract fast paths,
causing us to disregard further GPU damage and losing track of render
coherency.
Reported-by: Roman Jarosz <kedgedev@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41769
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 14:24:04 +01:00
46f97127c2
snb,ivb: Workaround unknown blitter death
...
The first workaround was a performance killing MI_FLUSH_DW after every
op. This workaround appears to be a stable compromise instead, only
requiring a redundant command after every BLT command with little
impact on throughput.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=27892
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=39524
Tested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 13:09:17 +01:00
71bf291e56
sna/gen3: Micro-optimise gen3_rectangle_begin()
...
We only need to emit the vbo description either at the beginning of a
new op (when the state may have changed) or after finishing a full vbo.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 12:53:19 +01:00
a032feb253
sna/gen3: Store floats_per_rect alongside floats_per_vertex
...
Moves a multiply out of the hot path.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 12:41:21 +01:00
2cbe79b85b
sna/gen3: Improve reduction of render operator to blt
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 11:57:07 +01:00
606e18bd36
configure: Add a check for pixman version
...
As SNA requires a fairly recent release and calls directly into pixman.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 10:50:33 +01:00
17acaf8e23
sna: Defer CPU fallback if fb calls into mi
...
If the fb routine does not access the pixel data directly, but instead
calls into an mi routine, we can defer the readback and possibly avoid
it.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-17 00:58:02 +01:00
c348b69f66
sna: Prefer to accelerate non-RENDER operations if already using the GPU
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
6fdc9e3fd3
sna: Simplify busy tracking by trusting the bo->gpu flag
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
30dff81cd6
sna/accel: Upload source to GPU for !copy CopyAreas
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
eefa925e3d
sna/accel: Fall-forward for handling a non-copy CopyArea to a dst gpu bo
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
01f41e624b
sna/gen2: Eliminate redundant diffuse and rectlist emission
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
d9e3dbffcb
sna/gen2: Fix fill-one-box
...
Lets only have one special gen2 value for the source channel pixel
colour and so remove the confusion and misrendering.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:24 +01:00
b89592b16e
sna/gen7: Add missing tail to 3DSTATE_SBE
...
I was a few NULs short of the declared amount.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-16 19:39:14 +01:00
33f17b996d
Remove a couple of trivial compile warnings for unused variables
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-14 12:01:34 +01:00
e5e27528ff
sna/blt: Add a pair of operator reductions for opaque fills
...
Extend the coverage of the BLT for a couple more composite operations,
though they be relatively infrequent.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-14 11:51:44 +01:00
3f5289b299
sna/trapezoids: Fallback if the composite operation is not supported
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-14 11:31:46 +01:00
3d79264ef8
sna/tiling: Hook up composite_box
...
Now used for spans and so triggering death by NULL function pointer on
older chipsets.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-14 11:31:00 +01:00
f5f3135e3c
sna: Ensure that the scanout is flushed before we sleep
...
So that we do not see any delayed rendering on i965+.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-14 10:40:39 +01:00
020d84d2d0
sna/accel: Fix sense of nothing-to-do due to clipping
...
A missing not due to a hasty conversion from an assert.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-13 22:32:00 +01:00
b7fd6906c4
sna/accel: Actually apply the clip to the glyph extents
...
References: https://bugs.freedesktop.org/show_bug.cgi?id=41718
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-13 17:52:14 +01:00
eb5f86b781
sna/accel: Correctly offset the damage intersection for glyph runs
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-13 17:52:10 +01:00
a4766026ad
sna/accel: Check for reloc overflow when adding a new glyph run
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-13 17:32:30 +01:00
b10850af41
sna/blt: Check for reloc overflow when initialising solid fills
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-13 17:09:25 +01:00
d5242565ef
sna/gen2: Trim some surplus state commands
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-12 23:28:57 +01:00
7c50205323
Remove vestigial includes from DRI1
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-12 20:36:53 +01:00
b6b3bb9b5e
sna: compare against the right array of cache buckets
...
Fortunately harmless today as there are the same number of
inactive/active buckets.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 17:14:18 +01:00
1f70095837
sna: And free the clip after creation for PolySegments
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 16:06:31 +01:00
92f4d978c8
sna: More micro-optimisation of messing around with clip regions
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 15:52:13 +01:00
57151f6547
sna: Micro-optimise checking for singular clip boxes
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 15:23:17 +01:00
823a4272c5
sna/gen3: Avoid RENDER/BLT context switch for fill boxes
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 13:51:41 +01:00
887361de17
sna: Enable single fill fast path for PolySegment
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 13:51:25 +01:00
721cf30e9e
sna/accel: If the data is already on the GPU, use it for the source
...
Fixes regression from 1ec6a0e277
2011-10-11 13:29:28 +01:00
15a4410cec
sna: use correct insertion point for sorting partials
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 13:28:30 +01:00
a9b53c4877
sna: Mark the spans render functions as fastcall
...
This reduces the amount of dancing required to call into the span
functions as we can pass the arguments in both the integer and floating
point registers.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 10:56:51 +01:00
208fa8e6b8
sna/trapezoid: Perform the NULL check for damage in the caller
...
Save the function call overhead in the common case.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 10:56:38 +01:00
5050fead0e
sna/gen3: avoid applying zero offset to common spans
...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
2011-10-11 10:54:53 +01:00
d0184b5909
snb: implement PIPE_CONTROL workaround
...
Sandybdrige requires an elaborate dance to flush caches without
hanging the gpu. See public docs Vol2Part1 1.7.4.1 PIPE_CONTROL
or the corrensponding code in mesa/kernel.
This (together with the corresponding patch for the kernel) seems to
fix the hangs in cairo-perf-traces I'm seeing on my snb machine.
v2: Incorporate review from Chris Wilson. For paranoia keep all three
PIPE_CONTROL cmds in the same batchbuffer to avoid upsetting the gpu.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2011-10-11 09:54:17 +02:00
Chris Wilson
Daniel Vetter