Fanin (merge) nodes require it's srcs to be "adjacent" in consecutive
scalar registers. Keep track of instruction neighbors in copy-
propagation step and avoid eliminating mov's which would cause an
instruction to need multiple distinct left and/or right neighbors.
This lets us not fall on our face when we encounter things like:
1: MOV TEMP[2], IN[0].xyzw
2: TEX OUT[0].xy, TEMP[2], SAMP[0], SHADOW2D
3: MOV TEMP[2].xy, IN[0].yxzz
4: TEX OUT[0].zw, TEMP[2], SAMP[0], SHADOW2D
5: END
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Always insert extra mov's for the tex coord into the fanin. This
simplifies things a bit, and avoids a scenario where multiple sam
instructions can have mutually exclusive input's to it's fanin, for
example:
1: TEX OUT[0].xy, IN[0].xyxx, SAMP[0], 2D
2: TEX OUT[0].zw, IN[0].yxxx, SAMP[0], 2D
The CP pass can always remove the mov's that are not actually needed,
so better to start out with too many mov's in the front end, than not
enough.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
dscis -> noscis
dbypass -> nobypass
a bit more consistant w/ nobin, etc. And IMO a bit more sensible names.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Kills get added to the outputs list, to ensure they get scheduled. But
they aren't *really* outputs so skip them in the header comment block.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
In order to test compiler changes more easily, spit out the assembled
shader with some header information so that we can know about
inputs/outputs more easily.
See: git://people.freedesktop.org/~robclark/ir3test
In ir3test we have a big collection of tgsi shaders and reference
ir3_compiler outputs. When making compiler changes, regenerate the
compiler outputs and feed to ir3test to compare the new vs reference
shader.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The last few dwords were skipped if the total number of dwords was not a
multiple of 4. Change the formatting for better readability.
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
In preparation of ARB_clip_control. Let the driver decide if
it supports pipe_rasterizer_state::clip_halfz being set to true.
v3:
Initially enable on ilo.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Froehlich <Mathias.Froehlich@web.de
This allows vc4_opt_cse.c to CSE-away operations involving the same
uniform values.
total instructions in shared programs: 37341 -> 36906 (-1.16%)
instructions in affected programs: 10233 -> 9798 (-4.25%)
total uniforms in shared programs: 10523 -> 10320 (-1.93%)
uniforms in affected programs: 2467 -> 2264 (-8.23%)
This saves a bunch of extra flushes when texsubimaging a whole texture
that's been used for rendering, or subdataing a whole BO. In particular,
this massively reduces the runtime of piglit texture-packed-formats (when
the probes have been moved out of the inner loop).
I'm going to be using VC4_DEBUG=shaderdb,norast to do shaderdb stats, but
when debugging regressions, I want to match shaderdb output to shader
disassembly.
This reverts commit 01e6371149.
Since then many Hyper-Z issues have been fixed or worked around.
Enable Hyper-Z by default so that we get enough feedback for the upcoming
mesa 10.4 release.
If you have issues with Hyper-Z try to disable Hyper-Z using the enviroment
variable R600_DEBUG=nohyperz and please report the issue on the bugtracker.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75011
See also: https://bugs.freedesktop.org/show_bug.cgi?id=75112
Signed-off-by: Andreas Boll <andreas.boll.dev@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
fd_bo_cpu_prep() doesn't realize the bo is already referenced in
unflushed cmdstream. It could be made to do so (but would have to be
implemented twice, ie. both for msm and kgsl). But we still can't do
the expected thing if the caller isn't using _NOSYNC. Because of the
way the tiling works, we need to build quite a bit of cmdstream at flush
time, which is not possible to do at the libdrm level.
So rather than trying to make fd_bo_cpu_prep() smarter than it can
possibly be, just *always* discard and reallocate if the
PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE flag is set.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
We'll need to update gallivm for the interface changes in LLVM 3.6, and
the fewer the number of older LLVM versions we support the less hairy that
will be.
As consequence HAVE_AVX define can disappear. (Note HAVE_AVX meant
whether LLVM version supports AVX or not. Runtime support for AVX is
always checked and enforced independently.)
Verified llvmpipe builds and runs with with LLVM 3.3, 3.4, and 3.5.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
We don't have a scissor enable bit in hw, so when a raster state change
results in scissor enable bit changing, we need to also mark scissor
state as dirty.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The optimization of avoiding restore (mem2gmem) if there was a clear
falls down a bit if you don't have a fullscreen scissor. We need to
make the decision logic a bit more clever to keep track of *what* was
cleared, so that we can (a) completely skip mem2gmem if entire buffer
was cleared, or (b) skip mem2gmem on a per-tile basis for tiles that
were completely cleared.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
The values are hardcoded in the LLVM backend, but the TGSI definitions are
going to be changed with tessellation, e.g. TGSI_PROCESSOR_COMPUTE will be
increased by 2.
We'll use VS for LS and HS, because there's nothing special about them
from the LLVM backend point of view, even though the hardware side is
different. We do the same for ES.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
I'll need indexed loads without the meta data flag for tessellation later.
Also rename load_const to buffer_load_const to distinguish it from indexed
const loads.
v2: add comments
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
With 5 shader stages and various combinations of enabled and disabled shaders,
the maximum number of outputs in one shader doesn't have to be equal to
the maximum number of inputs in the following shader.
v2: return 32 for softpipe and llvmpipe
If the writemask doesn't compress, then we want to put in the uncompressed
writemask, not the compressed writemask failure value (all-on).
Fixes glean's stencil2 and fbo-clear-formats on stencil.
It seems like the hardware is unhappy if we execute a kill instruction
prior to last input (ei). Probably the shader thread stops executing
and the end-input flag is never set.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
If app only updates (for example) vertex uniforms, it would be nice to
only re-emit those and not also frag uniforms. Means we need to mark
the first frag shader const buffer dirty after a clear.
Signed-off-by: Rob Clark <robclark@freedesktop.org>