Note that these limits are exact, not a "precision is at least x",
as texel coords also get snapped to a multiple of this step size
before filtering.
This fixes CTS tests
dEQP-VK.texture.explicit_lod.2d.sizes.31x55_nearest_linear_mipmap_nearest_repeat
dEQP-VK.texture.explicit_lod.2d.sizes.57x35_nearest_linear_mipmap_nearest_repeat
Fixes: f4e499ec79 "radv: add initial non-conformant radv vulkan driver"
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109151
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
These days, we have two sampler lowering passes. The newer one,
gl_nir_lower_samplers_as_deref, is used by radeonsi. It rewrites
variables to drop structures out of sampler deref chains, to make
life simpler. It then sets var->data.binding for non-bindless
sampler and image variables based on the GL uniform storage's
opaque index values.
The older one converts sampler deref chains (nir_tex_src_texture_deref)
to a numerical offset (nir_tex_src_texture_offset). It also stores the
constant-valued portion of that number in tex->texture_index, making
life really simple for drivers that don't support indirects. It too
pokes at GL uniform storage's opaque index values.
Logically, we can do the first pass (simplify derefs, set bindings)
then the second (turn derefs to offsets, set texture_index). This
patch does exactly that, eliminating some redundancy (only one pass
has to poke at GL uniform storage), and gaining proper var->data.binding
values for drivers using the full lowering.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
We recurse to remove structures, and at each step, re-modify the
resulting type for our link in the deref chain. For arrays, the
result of recursion is the new underlying type - so we wrap it with
the array dimensionality again. For structs, we want to simply use
the new underlying type, skipping the struct altogether.
The correct way to do this is to do nothing at all. Previously, we
had reset type to next->type, which is the /old/ field type, not the
new field type we obtained by recursing. This undid our recursive work.
Fixes about 338 tests with nested structs, such as:
dEQP-GLES2.functional.uniform_api.value.initial.get_uniform.nested_structs_arrays.sampler2D_samplerCube_fragment
Note that currently only radeonsi uses this pass, and NIR support is
disabled there by default, so the breakage was likely not seen by most
people. The next commit uses this pass for more drivers, so this fix
prevents regressions from that change.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
unused and gcc complains about strncpy. (from what I can see because
strncpy does not leave a 0 byte on truncate. That said we don't use
it so this does not fix a real bug).
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
We do the ImageFormatProperties check already, and rejecting an usage
flag when both ImageFormatProperties and the WSI (which is Android)
support it is not allowed.
Intel does support storage for some of the support WSI formats, such
as R8G8B8A8_UNORM, and looking at the ISL_SURF_USAGE_DISABLE_AUX_BIT,
the imported images do not have any form of compression that would
prevent this fix.
v2: Also consider STORAGE bit for Gralloc usage bits.
(From Kevin Strasser <kevin.strasser@intel.com>)
Fixes: 053d4c328f "anv: Implement VK_ANDROID_native_buffer (v9)"
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
We started using it in the btoi paths for r32g32b32, and the LLVM IR
checker will complain about it because we end up with intrinsics with
the wrong type extension in the name.
Fixes: 593996bc02 ("radv: implement buffer to image operations for R32G32B32")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Some of the status variables in the compiler are only used in asserts
and thus may be unused in release builds. Annotate them accordingly
to avoid 'unused but set' warnings from the compiler.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Take into account the render target format when checking if the color
mask affects all channels of the RT. This allows to enable full
overwrite in a few cases where a non-alpha format is used.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
The functional change here is moving the nir_lower_io_to_scalar_early()
calls inside st_nir_link_shaders() and moving the st_nir_opts() call
after the call to nir_lower_io_arrays_to_elements().
This fixes a bug with the following piglit test due to the current code
not cleaning up dead code after we lower arrays. This was causing an
assert in the new duplicate varyings link time opt introduced in
70be9afccb.
tests/spec/glsl-1.10/execution/vsfs-unused-array-member.shader_test
Moving the nir_lower_io_to_scalar_early() calls also allows us to tidy
up the code a little and merge some loops.
Reviewed-by: Eric Anholt <eric@anholt.net>
Even without any clever optimization on the unpack operations, this gives
us a useful value for the channels read field, which we can use to avoid
ldtmu instructions to the no-op register.
instructions in affected programs: 890712 -> 881974 (-0.98%)
I've been doing this in the nir-to-vir and nir-to-qir backends of v3d and
vc4, but nir could potentially do some useful stuff for us (like avoiding
unpack/repacks) if we give it the information.
v2: Skip lowering for txs/query_levels
v3: Fix a crash on old-style shadow
v4: Rename to tex_packing, use nir_format_unpack_sint/uint helpers, pack
the enum.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
In 92eb5bbc68 we attempted to avoid copying clear colors whenever
we weren't doing a resolve. However, this broke MSAA resolves because
we need the clear color in the source. This patch makes blorp much more
conservative such that it only avoids the clear color copy if either
aux_usage == NONE or it's explicitly doing a fast-clear.
Fixes: 92eb5bbc68 "intel/blorp: Only copy clear color when doing..."
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107728
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
We can pull a whole vector in a single indirect load. This saves a bunch
of round-trips to the TMU, instructions for setting up multiple loads,
references to the UBO base in the uniforms, and apparently manages to
reduce register pressure as well.
instructions in affected programs: 3086665 -> 2454967 (-20.47%)
uniforms in affected programs: 919581 -> 721039 (-21.59%)
threads in affected programs: 1710 -> 3420 (100.00%)
spills in affected programs: 596 -> 522 (-12.42%)
fills in affected programs: 680 -> 562 (-17.35%)
Improves 3dmmes performance by 2.29312% +/- 0.139825% (n=5)
In the process of adding support for SSBOs and CS shared vars, I ended up
needing a helper function for doing TMU general ops. This helper can be
that starting point, and saves us a bunch of round-trips to the TMU by
loading a vector at a time.
I noticed that a VS I was debugging was missing all of its output stores
-- outputs_written was for POS, VAR0, VAR3, while the shader's variables
were POS, VAR9, and VAR12. I'm not sure what outputs_written is supposed
to be doing here, but we can just walk the declared variables and avoid
both this bug and the emission of extra stvpms for less-than-vec4
varyings.
When copy_prop_vars also took care of dead write handling, intrin was
used as part of store_to_entry. Now it isn't, so this assignment
isn't used really used. Add a comment clarifying what happens to
intrin.
Fixes: 4dfa7adc10 "nir: Remove handling of dead writes from copy_prop_vars"
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Even with the previous commit, hangs are still happening. The problem
there is that the VF cache invalidate do happen immediately without
waiting for previous rendering to complete. What happens is that we
invalidate the cache the moment the PIPE_CONTROL is parsed but we
still have old rendering in the pipe which continues to pull data into
the cache with the old high address bits. The later rendering with the
new high address bits then doesn't have the clean cache that it
expects/needs.
v2: Update commit message/explanation with Jason's
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Fixes: a363bb2cd0 ("i965: Allocate VMA in userspace for full-PPGTT systems.")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109072
Hopefully we can kick start the revolution and other distros will start
providing them as well :)
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
No functional change as the socket name is the same,
just removing the double definition of the path.
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Signed-off-by: Jakob Bornecrantz <jakob@collabora.com>
A 2d-array texture (for example), should get the # of array elements
from box->depth, rather than depth0 which is minified.
Fixes dEQP-GLES3.functional.shaders.texture_functions.texture.sampler2darray_bias_float_fragment
with tiled textures.
Reported-by: Kristian H. Kristensen <hoegsberg@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
If we hit the memcpy() path for copy_region(), that will try to do a
transfer_map(), which goes badly for blits to/from staging triggered
by transfer_map() or transfer_unmap().
We could possibly add fd_blit2() which has allow_transfer_map param,
and call that for staging blits. But I'm not really sure if trying
the blit via copy_region() is very useful. At least for newer gens
that implement fd_context::blit(), it probably isn't.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Switch over to using fd_context::blit(), in the same way that a5xx does.
The previous patch wires fd_resource_copy_region() up to the blitter so
a6xx no longer needs to bypass the core layer to accelerate this.
Signed-off-by: Rob Clark <robdclark@gmail.com>
First step to unify the way fd5 and fd6 blitter works. Currently a6xx
bypasses the blit API in order to also accelerate resource_copy_region()
But this approach can lead to infinite recursion:
#0 fd_alloc_staging (ctx=0x5555936480, rsc=0x7fac485f90, level=0, box=0x7fbab29220) at ../src/gallium/drivers/freedreno/freedreno_resource.c:291
#1 0x0000007fbdebed04 in fd_resource_transfer_map (pctx=0x5555936480, prsc=0x7fac485f90, level=0, usage=258, box=0x7fbab29220, pptrans=0x7fbab29240) at ../src/gallium/drivers/freedreno/freedreno_resource.c:479
#2 0x0000007fbe5c5068 in u_transfer_helper_transfer_map (pctx=0x5555936480, prsc=0x7fac485f90, level=0, usage=258, box=0x7fbab29220, pptrans=0x7fbab29240) at ../src/gallium/auxiliary/util/u_transfer_helper.c:243
#3 0x0000007fbde2dcb8 in util_resource_copy_region (pipe=0x5555936480, dst=0x7fac485f90, dst_level=0, dst_x=0, dst_y=0, dst_z=0, src=0x7fac47c780, src_level=0, src_box_in=0x7fbab2945c) at ../src/gallium/auxiliary/util/u_surface.c:350
#4 0x0000007fbdf2282c in fd_resource_copy_region (pctx=0x5555936480, dst=0x7fac485f90, dst_level=0, dstx=0, dsty=0, dstz=0, src=0x7fac47c780, src_level=0, src_box=0x7fbab2945c) at ../src/gallium/drivers/freedreno/freedreno_blitter.c:173
#5 0x0000007fbdf085d4 in fd6_resource_copy_region (pctx=0x5555936480, dst=0x7fac485f90, dst_level=0, dstx=0, dsty=0, dstz=0, src=0x7fac47c780, src_level=0, src_box=0x7fbab2945c) at ../src/gallium/drivers/freedreno/a6xx/fd6_blitter.c:587
#6 0x0000007fbde2f3d0 in util_try_blit_via_copy_region (ctx=0x5555936480, blit=0x7fbab29430) at ../src/gallium/auxiliary/util/u_surface.c:864
#7 0x0000007fbdec02c4 in fd_blit (pctx=0x5555936480, blit_info=0x7fbab29588) at ../src/gallium/drivers/freedreno/freedreno_resource.c:993
#8 0x0000007fbdf08408 in fd6_blit (pctx=0x5555936480, info=0x7fbab29588) at ../src/gallium/drivers/freedreno/a6xx/fd6_blitter.c:546
#9 0x0000007fbdebdc74 in do_blit (ctx=0x5555936480, blit=0x7fbab29588, fallback=false) at ../src/gallium/drivers/freedreno/freedreno_resource.c:129
#10 0x0000007fbdebe58c in fd_blit_from_staging (ctx=0x5555936480, trans=0x7fac47b7e8) at ../src/gallium/drivers/freedreno/freedreno_resource.c:326
#11 0x0000007fbdebea38 in fd_resource_transfer_unmap (pctx=0x5555936480, ptrans=0x7fac47b7e8) at ../src/gallium/drivers/freedreno/freedreno_resource.c:416
#12 0x0000007fbe5c5c68 in u_transfer_helper_transfer_unmap (pctx=0x5555936480, ptrans=0x7fac47b7e8) at ../src/gallium/auxiliary/util/u_transfer_helper.c:516
#13 0x0000007fbde2de24 in util_resource_copy_region (pipe=0x5555936480, dst=0x7fac485f90, dst_level=0, dst_x=0, dst_y=0, dst_z=0, src=0x7fac47b8e0, src_level=0, src_box_in=0x7fbab2997c) at ../src/gallium/auxiliary/util/u_surface.c:376
#14 0x0000007fbdf2282c in fd_resource_copy_region (pctx=0x5555936480, dst=0x7fac485f90, dst_level=0, dstx=0, dsty=0, dstz=0, src=0x7fac47b8e0, src_level=0, src_box=0x7fbab2997c) at ../src/gallium/drivers/freedreno/freedreno_blitter.c:173
#15 0x0000007fbdf085d4 in fd6_resource_copy_region (pctx=0x5555936480, dst=0x7fac485f90, dst_level=0, dstx=0, dsty=0, dstz=0, src=0x7fac47b8e0, src_level=0, src_box=0x7fbab2997c) at ../src/gallium/drivers/freedreno/a6xx/fd6_blitter.c:587
...
Instead rework the API to push the fallback back to core code, so that
we can rework resource_copy_region() to have it's own fallback path,
and then finally convert fd6 over to work in the same way.
This also makes ctx->blit() optional, and cleans up some unnecessary
callers.
Signed-off-by: Rob Clark <robdclark@gmail.com>
For multi-pass rendering, it is common to keep the same depth buffer
from previous pass, to discard geometry that would be hidden by later
draws. In the later passes with depth-test enabled, but depth-write
disabled, there is no reason to do gmem2mem resolve.
TODO probably do something similar for stencil.. although stencil
buffer isn't used as commonly these days
Signed-off-by: Rob Clark <robdclark@gmail.com>
After trying multiple times to merge if-statements with phis
between them I've come to the conclusion that it cannot be done
without regressions. The problem is for some shaders we end up
with a whole bunch of phis for the merged ifs resulting in
increased register pressure.
So this patch just merges ifs that have no phis between them.
This seems to be consistent with what LLVM does so for radeonsi
we only see a change (although its a large change) in a single
shader.
Shader-db results i965 (SKL):
total instructions in shared programs: 13098176 -> 13098152 (<.01%)
instructions in affected programs: 1326 -> 1302 (-1.81%)
helped: 4
HURT: 0
total cycles in shared programs: 332032989 -> 332037583 (<.01%)
cycles in affected programs: 60665 -> 65259 (7.57%)
helped: 0
HURT: 4
The cycles estimates reported by shader-db for i965 seem inaccurate
as the only difference in the final code is the removal of the
redundent condition evaluations and jumps.
Also the biggest code reduction (~7%) for radeonsi was in a tomb
raider tressfx shader but for some reason this does not get merged
for i965.
Shader-db results radeonsi (VEGA):
Totals from affected shaders:
SGPRS: 232 -> 232 (0.00 %)
VGPRS: 164 -> 164 (0.00 %)
Spilled SGPRs: 59 -> 59 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 14584 -> 13520 (-7.30 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 13 -> 13 (0.00 %)
Wait states: 0 -> 0 (0.00 %)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Jason Ekstrand