Tom Stellard [Tue, 17 May 2016 18:10:51 +0000 (18:10 +0000)]
radeon/llvm: Use alloca instructions for larger arrays
We were storing arrays in vectors, which was leading to some really bad
spill code for large arrays. allocas instructions are a better fit for
arrays and LLVM optimizations are more geared toward dealing with
allocas instead of vectors.
For arrays that have 16 or less 32-bit elements, we will continue to use
vectors, because this will force LLVM to store them in registers and
use indirect registers, which is usually faster for small arrays.
In the future we should use allocas for all arrays and teach LLVM
how to store allocas in registers.
This fixes the piglit test:
spec/glsl-1.50/execution/geometry/max-input-component
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tom Stellard [Tue, 17 May 2016 14:48:42 +0000 (14:48 +0000)]
radeon/llvm: Add helpers for loading and storing data from arrays.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tom Stellard [Tue, 17 May 2016 17:51:00 +0000 (17:51 +0000)]
radeon/llvm: Remove uses_temp_indirect_addressing() function
bld->indirect_files is never set, so this function always returns false.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Emil Velikov [Wed, 6 Jul 2016 15:18:21 +0000 (16:18 +0100)]
anv: vulkan: remove the anv_device.$(OBJEXT) rule
Atm the actual rule will expand to foo.o which is used for static
libraries only.
Thus the automake manual recommendation [to use OBJEXT] won't help us,
since since we're working with a shared library.
Thus let's 'demote' the file and add it back to BUILT_SOURCES. This will
manage all the complexity for us, at the (existing expense) of working
only with the all, check and install targets.
The crazy (why the issue was hard to spot):
If the dependencies (.deps/*.Plo) are already created one can alter the
anv_device.$(OBJEXT) line and/or nuke it all together. That won't lead
to any warnings/issues, even though the Makefile is regenerated.
Moral of the story:
Always rm -rf top_builddir or don't resolve the dependencies manually
and use BUILT_SOURCES.
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96825
Fixes: d7a604c3f7a ("anv: use cache uuid based on the build timestamp.")
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Rob Clark [Tue, 5 Jul 2016 19:41:26 +0000 (15:41 -0400)]
vbo: fix attr reset
In
bc4e0c4 (vbo: Use a bitmask to track the active arrays in vbo_exec*.)
we stopped looping over all the attributes and resetting all slots.
Which exposed an issue in vbo_exec_bind_arrays() for handling GENERIC0
vs. POS.
Split out a helper which can reset a particular slot, so that
vbo_exec_bind_arrays() can re-use it to reset POS.
This fixes an issue with 0ad (and possibly others).
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Rob Clark [Sat, 2 Jul 2016 12:02:51 +0000 (08:02 -0400)]
list: fix list_replace() for empty lists
Before, it would happily copy list_head next/prev (ie. pointer to the
*from* list_head), leaving things in a confused state and causing much
mayhem.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Rob Clark [Sun, 26 Jun 2016 17:56:42 +0000 (13:56 -0400)]
gallium: un-inline pipe_surface_desc
Want to re-use this struct, so un-inline it.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Rob Clark [Sat, 11 Jun 2016 13:21:10 +0000 (09:21 -0400)]
gallium/util: make util_copy_framebuffer_state(src=NULL) work
Be more consistent with the other u_inlines util_copy_xyz_state()
helpers and support NULL src.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Fri, 1 Jul 2016 14:18:34 +0000 (16:18 +0200)]
winsys/amdgpu: avoid flushed depth when possible
If a depth/stencil texture has no mipmaps, we can always get a layout that is
compatible with DB and TC.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Thu, 30 Jun 2016 18:04:54 +0000 (20:04 +0200)]
gallium/radeon: add depth/stencil_adjusted output to surface computation
This fixes a rare bug with stencil texturing -- seen on Polaris and Tonga,
though it's basically a function of the memory configuration so could affect
other parts as well.
Fixes piglit "unaligned-blit * stencil downsample" and various
"fbo-depth-array *stencil*" tests.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Thu, 30 Jun 2016 15:22:54 +0000 (17:22 +0200)]
gallium/radeon: allocate only the required plane for flushed depth
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Wed, 29 Jun 2016 12:30:37 +0000 (14:30 +0200)]
radeonsi: decompress to flushed depth texture when required
v2: s/dirty_level_mask/stencil_dirty_level_mask/ in stencil case
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Thu, 30 Jun 2016 14:25:39 +0000 (16:25 +0200)]
radeonsi: extract DB->CB copy logic into its own function
Also clean up some of the looping.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Wed, 29 Jun 2016 19:57:00 +0000 (21:57 +0200)]
radeonsi: sample from flushed depth texture when required
Note that this has no effect yet. A case where can_sample_z/s can be false
in radeonsi will be added in a later patch.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Thu, 30 Jun 2016 09:26:13 +0000 (11:26 +0200)]
gallium/radeon: replace is_flushing_texture with db_compatible
This is a left-over of when I considered generalizing the separate stencil
support. I do prefer the new name since it emphasizes what flushing vs.
non-flushing means from a functional point-of-view, namely special handling
of the texture format.
v2: adjust r600_init_color_surface as well
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Wed, 29 Jun 2016 19:56:42 +0000 (21:56 +0200)]
gallium/radeon: add can_sample_z/s flags for textures
v2: adjust r600_init_color_surface as well
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Fri, 1 Jul 2016 09:37:59 +0000 (11:37 +0200)]
radeonsi: correctly mark levels of 3D textures as fully decompressed
Account for the fact that max_layer is minified for higher levels.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Wed, 29 Jun 2016 12:07:44 +0000 (14:07 +0200)]
gallium/radeon/winsyses: remove unused stencil_offset
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Tue, 28 Jun 2016 07:40:26 +0000 (09:40 +0200)]
gallium/radeon: remove redundant null-pointer check
v2: keep using r600_texture_reference
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Mon, 27 Jun 2016 14:59:55 +0000 (16:59 +0200)]
gallium/radeon: print StencilLayout only once
It is the same for all levels.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Nicolai Hähnle [Wed, 22 Jun 2016 18:05:12 +0000 (20:05 +0200)]
gallium/radeon: flush stdout after printing texture information
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Ilia Mirkin [Fri, 1 Jul 2016 23:10:36 +0000 (19:10 -0400)]
glsl: don't try to lower non-gl builtins as if they were gl_FragData
If a shader has an output array, it will get treated as though it were
gl_FragData and rewritten into gl_out_FragData instances. We only want
this to happen on the actual gl_FragData and not everything else.
This is a small part of the problem pointed out by the below bug.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96765
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "11.2 12.0" <mesa-stable@lists.freedesktop.org>
Ian Romanick [Wed, 29 Jun 2016 22:17:16 +0000 (15:17 -0700)]
glsl: Document and enforce restriction on type values
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Ian Romanick [Tue, 14 Jun 2016 21:38:04 +0000 (14:38 -0700)]
glsl: Pack integer and double varyings as flat even if interpolation mode is none
v2: Also update varying_matches::compute_packing_class(). Suggested by
Timothy Arceri.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96358
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Cc: Gregory Hainaut <gregory.hainaut@gmail.com>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Ian Romanick [Fri, 10 Jun 2016 15:01:30 +0000 (08:01 -0700)]
mesa: Strip arrayness from interface block names in some IO validation
Outputs from the vertex shader need to be able to match
per-vertex-arrayed inputs of later stages. Acomplish this by stripping
one level of arrayness from the names and types of outputs going to a
per-vertex-arrayed stage.
v2: Add missing checks for TESS_EVAL->GEOMETRY. Noticed by Timothy
Arceri.
v3: Use a slightly simpler stage check suggested by Ilia.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96358
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Cc: Gregory Hainaut <gregory.hainaut@gmail.com>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Charmaine Lee [Tue, 28 Jun 2016 23:09:59 +0000 (16:09 -0700)]
svga: avoid emitting redundant DXSetRenderTargets command
Tested with Lightsmark2008, MTT piglit, glretrace, conform.
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Leo Liu [Wed, 29 Jun 2016 15:48:53 +0000 (11:48 -0400)]
radeon/vce: update encRefPic addr and array mode to tiled
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Leo Liu [Mon, 4 Jul 2016 15:29:28 +0000 (11:29 -0400)]
radeon/vce: increase cpb height alignment
Height should be aligned with 2 macroblocks, thus making safer
for tiled mode
Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Iago Toral Quiroga [Tue, 5 Jul 2016 12:06:37 +0000 (14:06 +0200)]
i965: Remove trailing whitespace
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Iago Toral Quiroga [Tue, 5 Jul 2016 12:05:40 +0000 (14:05 +0200)]
i965: Make inline function static
Without this the i965 driver fails to load.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Emil Velikov [Fri, 24 Jun 2016 15:35:05 +0000 (16:35 +0100)]
anv: install the intel_icd.json to ${datarootdir} by default
As mentioned by the spec (and used by Archlinux and Debian) default to
${datarootdir} as opposed to ${sysconfdir} for the default location.
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Emil Velikov [Fri, 1 Jul 2016 16:17:13 +0000 (17:17 +0100)]
swr: automake: don't ship LLVM version specific generated sources
Otherwise things will fail to build, if the builder is using another
version of LLVM.
v2: annotate all the dependencies of builder_gen.h
v3: clean the generated files as needed
v4: comment cleanups (Tim)
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Tested-by: Tim Rowley <timothy.o.rowley@intel.com>
Tested-by: Chuck Atkins <chuck.atkins@kitware.com> (v2)
Reported-by: Chuck Atkins <chuck.atkins@kitware.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Emil Velikov [Tue, 21 Jun 2016 17:05:23 +0000 (18:05 +0100)]
automake: don't mandate git_sha1.h/MESA_GIT_SHA1
It has proven subtle to get it right both from the build side POV (see
commit list below) and builders due to their varying workflows.
Furthermore it does not fully fulfil the reason why it was enforced -
to detect uniqueness between different builds, in order to distinguish
and invalidate Vulkan/GL caches.
With that having a much better solution (previous commit) we can drop
this solution.
This effectively reverts the following commits:
359d9dfec33 ("mesa: automake: add directory prefix for git_sha1.h")
2c424e00c39 ("mesa: automake: ensure that git_sha1.h.tmp has the right
attributes")
b7f7ec78435 ("mesa: automake: distclean git_sha1.h when building OOT")
8229fe68b5d ("automake: get in-tree `make distclean' working again.")
Cc: Timo Aaltonen <tjaalton@debian.org>
Cc: Haixia Shi <hshi@chromium.org>
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Emil Velikov [Fri, 1 Jul 2016 16:16:30 +0000 (17:16 +0100)]
anv: automake: indent with tabs and not spaces
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Emil Velikov [Tue, 21 Jun 2016 16:43:35 +0000 (17:43 +0100)]
anv: use cache uuid based on the build timestamp.
Do not rely on the git sha1:
- its current truncated form makes it less unique
- it does not attribute for local (Vulkand or otherwise) changes
Use a timestamp produced at the time of build. It's perfectly unique,
unless someone explicitly thinkers with their system clock. Even then
chances of producing the exact same one are very small, if not zero.
v2: Remove .tmp rule. Its not needed since we want for the header to be
regenerated on each time we call make (Eric).
v3:
- Honour SOURCE_DATE_EPOCH, to make the build reproducible (Michel)
- Replace the generated header with a define, to prevent needless
builds on consecutive `make' and/or `make install' calls. (Dave)
v4:
- Keep the timestamp generation at make time. (Jason)
v5:
- Ensure that file is regenerated on incremental builds.
Cc: Michel Dänzer <michel@daenzer.net>
Cc: Dave Airlie <airlied@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Emil Velikov [Tue, 21 Jun 2016 17:21:54 +0000 (18:21 +0100)]
clover: conditionally use MESA_GIT_SHA1
Considering how hard/annoying it was for many peoples' workflow to
properly generate the macro, it will be demoted to conditionally
available with follow-up commits.
v2: Kill off gracious blank line (Vedran).
Cc: mesa-stable@lists.freedesktop.org
Cc: Vedran Miletić <vedran@miletic.net>
Cc: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com> (v1)
Reviewed-by: Vedran Miletić <vedran@miletic.net>
Timothy Arceri [Sun, 3 Jul 2016 09:16:23 +0000 (19:16 +1000)]
mesa: stop copying SamplerUnits twice
The call to _mesa_update_shader_textures_used() already takes
care of copying for us.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Timothy Arceri [Fri, 1 Jul 2016 04:00:23 +0000 (14:00 +1000)]
mesa: make attribute binding message more useful
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Timothy Arceri [Tue, 5 Jul 2016 07:02:46 +0000 (17:02 +1000)]
i965: make more effective use of SamplersUsed
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Timothy Arceri [Sat, 2 Jul 2016 10:02:47 +0000 (20:02 +1000)]
glsl: stop allocating memory for UBOs during linking
This just stops counting and assigning a storage location for
these uniforms, the count is only used to create the uniform storage.
These uniform types don't use this storage.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Timothy Arceri [Sat, 2 Jul 2016 03:36:09 +0000 (13:36 +1000)]
glsl: mark link_uniform_blocks_are_compatible() as static
Missed this when doing
6d1a59d15b.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Timothy Arceri [Tue, 5 Jul 2016 08:40:30 +0000 (18:40 +1000)]
mesa: fix build error
Fix build error cased by
6a524c76f5.
Gregory Hainaut [Fri, 24 Jun 2016 08:07:02 +0000 (10:07 +0200)]
mesa: faster validation of sampler unit mapping for SSO
Code was inspired from _mesa_update_shader_textures_used
However unlike _mesa_update_shader_textures_used that only check for a single
stage, it will check all stages.
It avoids to loop on all uniforms, only active samplers are checked.
For my use case: high FS frequency switches with few samplers.
Perf event (relative to nouveau_dri.so) goes from 5.01% to 1.68% for
the _mesa_sampler_uniforms_pipeline_are_valid function.
Signed-off-by: Gregory Hainaut <gregory.hainaut@gmail.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Dave Airlie [Tue, 5 Jul 2016 00:22:21 +0000 (10:22 +1000)]
Revert "st/glsl_to_tgsi: don't increase immediate index by 1."
This reverts commit
27d456cc87a01998c6fe1dbf45937e2ca6128495.
DOH, what seems right and what is right with fp64 are always
two different things.
This regressed:
spec@arb_gpu_shader_fp64@shader_storage@layout-std140-fp64-mixed-shader
on radeonsi
Reported-by: Michel Dänzer <michel@daenzer.net>
Cc: "11.2 12.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Samuel Pitoiset [Mon, 4 Jul 2016 22:20:00 +0000 (00:20 +0200)]
nvc0/ir: rename NVE4_SU_INFO_XXX to NVC0_SU_INFO_XXX
While we are at it, fix a typo inside the comment which describes
what those constants are for.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Mon, 4 Jul 2016 22:08:03 +0000 (00:08 +0200)]
nvc0/ir: reset the base offset for indirect images accesses
In presence of an indirect image access, the base offset should be
zeroed because the stride will be computed twice. This is a pretty
rare situation but it can happen when tex.r > 0.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.2 12.0" <mesa-stable@lists.freedesktop.org>
Samuel Pitoiset [Mon, 4 Jul 2016 11:12:20 +0000 (13:12 +0200)]
gm107/ir: fix sign bit emission for FADD32I
When emitting OP_SUB, the sign bit for FADD and FADD32I is not
at the same position. It's at position 45 for FADD but 51 for FADD32I.
This fixes the following piglit test:
tests/spec/arb_fragment_program/fdo30337b.shader_test
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: <mesa-stable@lists.freedesktop.org>
Eric Anholt [Sun, 26 Jun 2016 02:49:07 +0000 (19:49 -0700)]
vc4: Regularize instruction emit macros
ALU0 didn't have the _dest variant, and ALU2 didn't unset the def the way
ALU1 did. This should make the ALU[012] macros much clearer, by moving
most of their contents to vc4_qir.c
Eric Anholt [Sun, 26 Jun 2016 01:45:28 +0000 (18:45 -0700)]
vc4: Enable dead CF elimination.
Now that we're about to start generating control flow in our NIR, we want
this in place. It optimizes things frequently in the CS, when the GL VS
has control flow that doesn't affect the vertex position.
Eric Anholt [Sat, 4 Jun 2016 00:09:14 +0000 (17:09 -0700)]
vc4: Optimize out redundant SF updates.
Tiny change on shader-db currently, but it will be important when we start
emitting a lot of SFs from the same variable as part of control flow
support.
total instructions in shared programs: 89463 -> 89430 (-0.04%)
instructions in affected programs: 1522 -> 1489 (-2.17%)
total estimated cycles in shared programs: 250060 -> 250015 (-0.02%)
estimated cycles in affected programs: 8568 -> 8523 (-0.53%)
Eric Anholt [Fri, 3 Jun 2016 21:36:04 +0000 (14:36 -0700)]
vc4: Move SF removal to a separate peephole pass.
The DCE pass is going to change significantly to handle control flow,
while we don't really need to change it for the SF handling. We also need
to add some more SF peephole optimization for SF updates generated by
control flow support.
No change on shader-db.
Eric Anholt [Fri, 3 Jun 2016 22:06:52 +0000 (15:06 -0700)]
vc4: DCE instructions with a NULL destination.
I'm going to add an optimization for redundant SF update removal, which
will just remove the SF and leave us (in many cases) with an instruction
with a NULL destination and no side effects. Rather than teaching that
pass whether the whole instruction can be removed, leave that
responsibility to this pass.
Eric Anholt [Fri, 3 Jun 2016 22:04:57 +0000 (15:04 -0700)]
vc4: Mark texturing setup instructions as having side effects.
We need to not DCE them even though they don't have a destination in QIR.
We also shouldn't relocate them in vc4_opt_vpm. Neither of these things
happen, but I'm about to make DCE consider instructions with a NULL
destination.
Eric Anholt [Sat, 2 Jul 2016 22:08:27 +0000 (15:08 -0700)]
vc4: Fix a pasteo in scheduling condition flag usage.
Noticed by code inspection. This hasn't been too big of a deal, because
our cond usages all start out as adder ops, either MOVs or the FTOI for Z
writes. MOVs *can* get converted to mul ops during scheduling, but
apparently we hadn't hit this.
Eric Anholt [Sun, 26 Jun 2016 02:39:05 +0000 (19:39 -0700)]
vc4: Drop the dead QIR_PACK() macro.
This isn't used since we switched to using the dst.pack field instead of
custom instructions.
Marek Olšák [Sat, 11 Jun 2016 17:57:40 +0000 (19:57 +0200)]
radeonsi: do compilation from si_create_shader_selector asynchronously
Main shader parts and geometry shaders are compiled asynchronously
by util_queue. si_create_shader_selector doesn't wait and returns.
si_draw_vbo(si_shader_select) waits for completion.
This has the best effect when shaders are compiled at app-loading time.
It doesn't help much for shaders compiled on demand, even though
VS+PS compilation should take as much as time as the bigger one of the two.
If an app creates more shaders, at most 4 threads will be used to compile
them.
Debug output disables this for shader stats to be printed in the correct
order.
(We could go even further and build variants asynchronously too, then emit
draw calls without waiting and emit incomplete shader states, then force IB
chaining to give the compiler more time, then sync the compilation at the IB
flush and patch the IB with correct shader states. This is great for
compilation before draw calls, but there are some difficulties such as
scratch and tess states requiring the compiler output, and an on-disk shader
cache will likely be a much better and simpler solution.)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Sat, 11 Jun 2016 17:32:53 +0000 (19:32 +0200)]
radeonsi: don't lock shader cache mutex during compilation
to allow multiple shaders to be compiled simultaneously.
ALso, shader-db can again use all 4 cores.
v2: Remove the pipe_mutex_unlock call in the error path.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
Marek Olšák [Sat, 11 Jun 2016 16:59:26 +0000 (18:59 +0200)]
radeonsi: separate the compilation chunk of si_create_shader_selector
The function interface is ready to be used by util_queue.
Also, si_shader_select_with_key can no longer accept si_context.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Sat, 11 Jun 2016 16:53:15 +0000 (18:53 +0200)]
radeonsi: move LLVMTargetMachineRef creation to a separate function
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 22:17:36 +0000 (00:17 +0200)]
gallium/radeon: add and use radeon_info::max_alloc_size (v2)
v2: - squashed the patches
- use INT_MAX
- clamp max_const_buffer_size
- check the DRM version in radeon
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Vedran Miletić <vedran@miletic.net>
Marek Olšák [Thu, 30 Jun 2016 22:10:15 +0000 (00:10 +0200)]
radeonsi: print LLVM IRs to ddebug logs
Getting LLVM IRs of hanging shaders have never been easier.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 29 Jun 2016 23:15:19 +0000 (01:15 +0200)]
radeonsi: enable string markers and record apitrace call numbers
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 00:11:04 +0000 (02:11 +0200)]
ddebug: add an option to dump info about a specific apitrace call
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 29 Jun 2016 23:53:04 +0000 (01:53 +0200)]
ddebug: implement pipe_context::generate_mipmap
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 29 Jun 2016 22:40:01 +0000 (00:40 +0200)]
ddebug: record and dump apitrace call numbers
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 29 Jun 2016 21:30:12 +0000 (23:30 +0200)]
ddebug: implement emit_string_marker
and remove some obsolete comments
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 18:01:26 +0000 (20:01 +0200)]
gallium/radeon: remove unused code - radeon_llvm_util.*
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 22 Jun 2016 20:24:52 +0000 (22:24 +0200)]
radeonsi: keep using v_rcp_f32 for division in future LLVM (v2)
This will be needed after some LLVM changes that haven't landed yet.
v2: - use LLVMIsConstant to fix an LLVM assertion failure.
LLVMSetMetadata doesn't work with constants.
- don't set float metadata as string
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 22:58:19 +0000 (00:58 +0200)]
radeonsi: remove an obsolete comment
It's not true.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 11:28:49 +0000 (13:28 +0200)]
radeonsi: don't interpolate colors if flatshading is enabled
use v_interp_mov for those
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 00:16:16 +0000 (02:16 +0200)]
radeonsi: enable the barycentric optimization in all cases
Handle the bc_optimize SGPR bit if both CENTER and CENTROID are enabled.
This should increase the PS launch rate for big primitives with MSAA.
Based on discussion with SPI guys.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 08:57:34 +0000 (10:57 +0200)]
radeonsi: compute only one set of interpolation (i,j) when MSAA is disabled
This should increase the PS launch rate for shaders using at least 2 pairs
of perspective (i,j) and same for linear.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 08:50:26 +0000 (10:50 +0200)]
radeonsi: split ps.prolog.force_persample_interp into persp and linear bits
This reduces the number of v_mov's in the prolog.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 30 Jun 2016 09:10:21 +0000 (11:10 +0200)]
radeonsi: don't dump the shader key for non-monolithic shaders early
It's always zero.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Jan Vesely [Sat, 2 Jul 2016 20:14:54 +0000 (16:14 -0400)]
r600g: Add double precision FMA ops
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96782
Fixes: 54c4d525da7c7fc1e103d7a3e6db015abb132d5d ("r600g: Enable FMA on chips that support it")
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Tested-by: James Harvey <lothmordor@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Francesco Ansanelli [Sun, 3 Jul 2016 19:03:04 +0000 (21:03 +0200)]
r600: fix duplicate 'const' declaration
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Topi Pohjolainen [Mon, 23 May 2016 06:22:01 +0000 (09:22 +0300)]
i965/urb: Allow blorp to record current settings
This makes it possible to skip urb re-configuration if the
subsequent renders agree with the settings.
Also allows blorp to allocate the maximun amount of vs entries
available. Core upload logic already knows how to calculate this.
Helps one synthetic benchmark.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Topi Pohjolainen [Wed, 18 May 2016 19:07:05 +0000 (22:07 +0300)]
i965/blorp/gen7+: Do not trigger push constant space reconfig
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Fri, 20 May 2016 08:15:35 +0000 (11:15 +0300)]
i965/blorp/gen7+: Stop trashing push constant allocation
Packet 3DSTATE_CONSTANT_PS is still emitted explicitly as ps stage
itself is enabled and hardware may try to prefetch constants from
the buffer. From the BSpec: 3D Pipeline - Windower -
3DSTATE_PUSH_CONSTANT_ALLOC_PS
"Specifies the size of the PS constant buffer. This value will
determine the amount of data the command stream can pre-fetch
before the buffer is full."
This is not possible on gen6. From the BSpec about 3DSTATE_CONSTANT_PS:
"This packet must be followed by WM_STATE."
Binding table emissions for stages other than PS can be now dropped,
they were only needed for the 3DSTATE_CONSTANT_XS to be effective:
From the BSpec:
"The 3DSTATE_CONSTANT_* command is not committed to the shader unit
until the corresponding (same shader) 3DSTATE_BINDING_TABLE_POINTER_*
command is parsed."
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Wed, 18 May 2016 19:01:17 +0000 (22:01 +0300)]
i965/blorp: Remove support for push constants
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 04:43:39 +0000 (07:43 +0300)]
i965/blorp: Use flat inputs instead of uniforms
v2 (Jason): Use LOAD_INPUT() macro
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Wed, 18 May 2016 04:30:00 +0000 (07:30 +0300)]
i965/blorp: Fix the size requirement for vertex elements
v2: Rebased as this is needed before flat inputs are enabled
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Topi Pohjolainen [Tue, 17 May 2016 13:41:23 +0000 (16:41 +0300)]
i965/blorp: Load tranformation coordinates as vec4
In preparation for loading as flat vertex input.
v2: Use LOAD_INPUT() macro
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 3 Jul 2016 07:19:25 +0000 (10:19 +0300)]
i965/blorp: Rename LOAD_UNIFORM to LOAD_INPUT
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Tue, 17 May 2016 06:27:49 +0000 (09:27 +0300)]
i965/blorp: Organize pixel kill and blend/scaled inputs into vec4s
In addition, as these are never used in parallel, add a few
assertions.
v2 (Jason): Skip some complexity by putting them into a union but
pad rectangle grid into a vec4 instead. Also keep the
LOAD_UNIFORM macro.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Lionel Landwerlin [Mon, 4 Jul 2016 17:06:58 +0000 (18:06 +0100)]
anv/wsi: create swapchain images using specified image usage
The image usage specified by the caller of vkCreateSwapchainKHR should be
passed onto the internal image creation. Otherwise the driver might later
crash when the user tries to use the image as a combined sampler even though
the creation was explicitly created with VK_IMAGE_USAGE_TRANSFER_SRC_BIT.
Leaving the previous VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT as this might be
expected even if the swapchain is created without any flag.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96791
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Indrajit Das [Fri, 27 May 2016 12:57:15 +0000 (18:27 +0530)]
radeon/uvd: fix overflow error while calculating bit stream buffer size
Reviewed-by: Christian König <christian.koenig@amd.com>
Topi Pohjolainen [Wed, 18 May 2016 16:25:33 +0000 (19:25 +0300)]
i965/blorp: Prepare for more than two vertex attributes
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 08:10:05 +0000 (11:10 +0300)]
i965/blorp: Tell vertex fetcher about flat inputs
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 05:33:16 +0000 (08:33 +0300)]
i965/blorp: Add support for flat input buffer
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Wed, 1 Jun 2016 09:27:56 +0000 (12:27 +0300)]
i965/blorp: Store input read mask
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Tue, 17 May 2016 12:44:39 +0000 (15:44 +0300)]
i965/blorp: Rename push constants to inputs
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 06:48:58 +0000 (09:48 +0300)]
i965/blorp: Use core vertex buffer state setup
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Thu, 23 Jun 2016 06:35:57 +0000 (09:35 +0300)]
i965/blorp: Split vertex data and element setup
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 06:42:39 +0000 (09:42 +0300)]
i965: Unify vertex buffer setup
On gen >= 8 one doesn't provide ending address but number of bytes
available. This is relative to the given offset.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Topi Pohjolainen [Sun, 15 May 2016 05:48:13 +0000 (08:48 +0300)]
i965/draw: Expose vertex buffer state setup
Also change the interface to use start and end offsets.
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Rob Clark [Sun, 3 Jul 2016 15:16:28 +0000 (11:16 -0400)]
freedreno: fix crash on smaller gpus and higher resolutions
Devices with smaller GMEM size need more tiles. On db410c at 2048x1152,
glmark2 shadow needed ~330 tiles for fullscreen. Lets bump it up to
512. (Maybe with MRT you could end up needing more, but at that point
things are probably going to be painfully slow.)
Signed-off-by: Rob Clark <robdclark@gmail.com>
Rob Clark [Sat, 25 Jun 2016 00:03:37 +0000 (20:03 -0400)]
i965: don't drop const initializers in vector splitting
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Rob Clark [Fri, 24 Jun 2016 18:28:51 +0000 (14:28 -0400)]
glsl: add driconf to zero-init unintialized vars
Some games are sloppy.. perhaps because it is defined behavior for DX or
perhaps because nv blob driver defaults things to zero.
So add driconf param to force uninitialized variables to default to zero.
This issue was observed with rust, from steam store. But has surfaced
elsewhere in the past.
Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Rob Clark [Fri, 24 Jun 2016 19:17:27 +0000 (15:17 -0400)]
freedreno/ir3: support glsl linking for cmdline compiler
For .vert/.frag, now multiple can be specified on the cmdline for
purposes of linking, and the last one specified is the one that is
fed into the ir3 backend (and dumped along the way if --verbose is
specified)
Without this, varyings in frag shaders would appear as undefined.
Signed-off-by: Rob Clark <robclark@freedesktop.org>