Increase the max allowed vector size from 256 to 512.
No piglit llvmpipe regressions running on avx2.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The original implementation allocated a new BO here, but we decided to
switch to intel_upload_space, which returns a reference to the current
upload BO. We accidentally kept the brw_bo_alloc, even though it's no
longer necessary - intel_upload_space will immediately unreference it,
causing us to allocate and immediately free a buffer.
Reviewed-by: Plamena Manolova <plamena.manolova@intel.com>
Section 6.3.2 of the GL 4.5 spec says:
"Any GL command which attempts to read from, write to, or change
the state of a buffer object may generate an INVALID_OPERATION error
if all or part of the buffer object is mapped ... However, only
commands which explicitly describe this error are required to do so.
If an error is not generated, such commands will have undefined
results and may result in GL interruption or termination."
Setting this flag allows us to skip walking over the buffer bindings
for every enabled vertex attribute (_mesa_all_buffers_are_unmapped).
Improves performance in GFXBench4's gl_driver2_off microbenchmark by
3.05797% +/- 0.709031% (n=33) on Apollolake.
This breaks KHR-*.draw_elements_base_vertex_tests.invalid_mapped_bos,
but that test is invalid and has been removed from the upstream CTS.
Reviewed-by: Eric Anholt <eric@anholt.net>
That requires a generated header that was rolled into a loop.
fixes: a47c525f32 ("meson: build glx")
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Previously buffer offsets were passed in explicitly as an offset, which
had to be added to the resource address. Now they are passed in via an
increased 'start' parameter. As a result, we were double-adding the
start offset in this kind of situation.
This condition was triggered by piglit's draw-elements test which has a
requisite glMultiDrawElements in combination with a small enough number
of vertices to go through the immediate push path.
Fixes: 330d0607ed ("gallium: remove pipe_index_buffer and set_index_buffer")
Reported-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
Commit 06bfb2d28f ("r600: fork and import gallium/radeon") broke the
Android build:
external/mesa3d/src/gallium/drivers/radeon/r600_pipe_common.c:43:10: fatal error: 'llvm-c/TargetMachine.h' file not found
^~~~~~~~~~~~~~~~~~~~~~~~
Update the Android makefiles so that drivers/radeon is only built when
radeonsi (and therefore LLVM) is enabled.
Fixes: 06bfb2d28f (r600: fork and import gallium/radeon)
Acked-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Rob Herring <robh@kernel.org>
As part of Treble project in Android O, all the device specific files have
to be located in a separate vendor partition. This is done by setting
LOCAL_PROPRIETARY_MODULE (the name is misleading). This change will not
break existing platforms without a vendor partition as it will just move
files to /system/vendor.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Rob Herring <robh@kernel.org>
ARB_enhanced_layouts allows multiple output variables to share the same
location - and these variables may not have the same sizes. For
example, consider these output variables:
// consume X/Y/Z components of 6 vectors
layout(location = 0) out vec3 a[6];
// consumes W component of the first vector
layout(location = 0, component = 3) out float b;
Looking at the first declaration, we see that VARYING_SLOT_VAR0 needs 24
components worth of space (vec3 padded out to a vec4, 4 * 6 = 24). But
looking at the second declaration, we would think that VARYING_SLOT_VAR0
needs only 4 components of space (a single float padded out to a vec4).
nir_setup_outputs() only considered the space requirements of the first
declaration it happened to see, so if 'float b' came first, it would
underallocate the output register space, causing brw_fs_validator.cpp
to assert fail about inst->dst.offset exceeding the register size.
Fixes Piglit's tests/spec/arb_enhanced_layouts/execution/component-layout/
vs-to-fs-array-interleave-single-location.shader_test.
Thanks to Tim Arceri for finding this bug and writing a test!
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
With the ssao demo from Vulkan demos:
radv/rx480: 440->440fps
anv/haswell: 24->34 fps
The demo does a 0->32 loop across a ubo with 32 members.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This assert was firing just running demos.
Jason said it should be this.
Fixes: 6c7720ed78 (anv/wsi: Allocate enough memory for the entire image)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This respects the size from the range setting like ssbo.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These paths are again 90% the same, consolidate them into
one.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These are 90% the same code, consolidate them into a couple of
common codepaths.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
One binding to bind them all, these are all the same thing.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
this adds automatic size support to the atomic buffer code,
but also realigns the code to act like the ubo/ssbo code.
v1.1:
add missing blank lines.
reindent one block properly.
check for NullBufferObj.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
KHR-GL45.shader_ballot_tests.ShaderBallotBitmasks has a MOV that hits
this validation path. MOVs don't have a src1 file, but calling
brw_inst_src1_type() was tripping on src1.file being BRW_IMMEDIATE_VALUE
and the hw_type being something invalid for immediates.
To work around this, just pretend src1 is src0 if there isn't a src1.
Fixes: 2572c2771d (i965: Validate "Special
Requirements for Handling Double Precision Data Types")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102680
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Fixes 'KHR-GL45.copy_image.functional' on Nouveau and i965.
v2: (by Kenneth Graunke)
Rewrite patch according to Jason Ekstrand's review feedback.
This makes it handle differing strides, which i965 needed.
Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Older kernels fail the va_op with this flag set. If the kernel
supports GFX9 usefully, it will also support this flag.
Fixes: e8d57802fe "radv/gfx9: allocate events from uncached VA space"
Reviewed-by: Dave Airlie <airlied@redhat.com>
Jason and I investigated several OpenGL CTS failures where the tests
bind the same texture for rendering and texturing, at the same time.
This has defined results as long as the reads happen before writes,
or the regions are non-overlapping. Normally, this just works out.
However, CCS can cause problems. If the shader is reading one set of
pixels, and writing to different pixels that are adjacent, they may end
up being covered by the same CCS block. So rendering may be writing a
CCS block, while the sampler is trying to read it. Corruption ensues.
Disabling CCS is unfortunate, but safe.
Fixes several KHR-GL45.texture_barrier.* subtests.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This lowers ffma to a * b + c.
This seems like it should keep Marek happiest, so
we'd never get to the fma instruction emission code.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
So it appears the Vulkan SPIR-V fma opcode can be equivalent to a
mad operation, and the fma hw opcode on AMD hw is issued like a double
opcode so is slower. Also the radeonsi stack does this.
This appears to improve performance on a number of games from Feral,
and thanks to Feral for noticing the problem.
I'm reposting this one as Marek indicated he thinks this is what
we should be doing on AMD hw.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The HW will halt when you hit a HALT packet, or when you hit the end
address. Tell CLIF if there's an end address is so that it can stop
correctly. (There was usually a 0 byte after the CL, so it would stop
anyway).
In order to keep early-Z from writing early in a discard shader, you need
to set the "modifies Z" bit in the shader state (which the new
prog_data.discards will indicate). Then, in the shader we do a TLB write
to make Z passthrough happen (the QPU result is ignored, so we use a NULL
source).
Tim Rowley
Kenneth Graunke
Dylan Baker
Ilia Mirkin
Rob Herring
Dave Airlie
Karol Herbst
Jason Ekstrand
Bas Nieuwenhuizen
Alex Smith
Eric Anholt