The kernel driver reports correct values now.
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
fixes to issues spotted by Emil Velikov:
- set ANV_TIMESTAMP corretly
- fix typo with VULKAN_GEM_FILES
v2: update to use Makefile.sources under vulkan
instead of having own
v3: update to changes to generate from vk.xml
(commit c7fc310)
v4: remove 'hw' relative path
cleanups, remove unnecessary cruft
review from Emil Velikov:
- move to vulkan folder
- remove timestamp gen, no longer necessary
- more cleanups
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
All the other calls to retrieve the attachment have been covered except
this one - return the proper error for attachment points that are valid
enums but out of bound for the driver.
Fixes GL45-CTS.geometry_shader.layered_fbo.fb_texture_invalid_attachment
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
It is not clear from the docs exactly how pipelined STATE_BASE_ADDRESS
actually is. We know from experimentation that we need to flush the
render cache prior to emitting STATE_BASE_ADDRESS and invalidate the
texture cache afterwards. The only thing the PRM says is that, on gen8+
we're supposed to invalidate the state cache after STATE_BASE_ADDRESS
but experimentation has indicated that doing so does nothing whatsoever.
Since we don't really know, let's do just a bit more flushing in the
hopes that this won't be a problem again. In particular:
1) Do a CS stall before we emit STATE_BASE_ADDRESS since we don't
really know whether or not it's pipelined.
2) Do a data cache flush in case what runs before STATE_BASE_ADDRESS
is a compute shader.
3) Invalidate the state and constant caches after STATE_BASE_ADDRESS
because the state may be getting cached there (we don't really know).
Reported-by: Mark Janes <mark.a.janes@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>
We had no good reason for *not* doing this on gen7 before but we didn't
know it was needed. Recently, when trying update to Vulkan CTS version
1.0.2 in our CI system, Mark discovered GPU hangs on Haswell that appear
to be STATE_BASE_ADDRESS related. This commit fixes them.
Reported-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>
This causes hangs on Broadwell if you try to render to it. I have no
idea how we managed to not hit this earlier.
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>
No f16 support as I'm not quite sure about alignment yet.
Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
The addition of Neon assembly breaks on arm64 builds because the assembly
syntax is different. For now, restrict Neon to ARMv7 builds.
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
clang throws an error on "%r2" and similar. I couldn't find any
documentation on what "%r?" is supposed to mean and I've never seen any
use like that as far as I remember. The parameter is supposed to be
cpu_stride and just %2/%3 should be sufficient.
There's no need for trailing ";" either, so remove those, too.
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
This prevents LLVM from using sext instructions for local memory offsets
and allows the backend to fold immediate offsets into the instruction.
This also prevents some incorrect code generation for ptrtoint and
inttoptr instructions.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
See "glsl: Rewrite atan2 implementation to fix accuracy and handling
of zero/infinity." for the rationale, but note that the instruction
count benefit discussed there is somewhat less important for the SPIRV
implementation, because the current code already emitted no control
flow instructions -- Still this saves us one hardware instruction per
scalar component on Intel SKL hardware.
Fixes the following Vulkan CTS tests on Intel hardware:
dEQP-VK.glsl.builtin.precision.atan2.highp_compute.scalar
dEQP-VK.glsl.builtin.precision.atan2.highp_compute.vec2
dEQP-VK.glsl.builtin.precision.atan2.highp_compute.vec3
dEQP-VK.glsl.builtin.precision.atan2.highp_compute.vec4
dEQP-VK.glsl.builtin.precision.atan2.mediump_compute.vec2
dEQP-VK.glsl.builtin.precision.atan2.mediump_compute.vec4
Note that most of the test-cases above expect IEEE-compliant handling
of atan2(±∞, ±∞), which this patch doesn't explicitly handle, so
except for the last two the test-cases above weren't expected to pass
yet. The reason they do is that the i965 back-end implementation of
the NIR fmin and fmax instructions is not quite GLSL-compliant (it
complies with IEEE 754 recommendations though), because fmin/fmax of a
NaN and a non-NaN argument currently always return the non-NaN
argument, which causes atan() to flush NaN to one and return the
expected value. The front-end should probably not be relying on this
behavior for correctness though because other back-ends are likely to
behave differently -- A follow-up patch will handle the atan2(±∞, ±∞)
corner cases explicitly.
v2: Fix up argument scaling to take into account the range and
precision of exotic FP24 hardware. Flip coordinate system for
arguments along the vertical line as if they were on the left
half-plane in order to avoid division by zero which may give
unspecified results on non-GLSL 4.1-capable hardware. Sprinkle in
some more comments.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This addresses several issues of the current atan2 implementation:
- Negative zero (and negative denorms which end up getting flushed to
zero) isn't handled correctly by the current implementation. The
reason is that it does 'y >= 0' and 'x < 0' comparisons to decide
on which side of the branch cut the argument is, which causes us to
return incorrect results (off by up to 2π) for very small negative
values.
- There is a serious precision problem for x values of large enough
magnitude introduced by the floating point division operation being
implemented as a mul+rcp sequence. This can lead to the quotient
getting flushed to zero in some cases introducing an error of over
8e6 ULP in the result -- Or in the most catastrophic case will
cause us to return NaN instead of the correct value ±π/2 for y=±∞
and x very large. We can fix this easily by scaling down both
arguments when the absolute value of the denominator goes above
certain threshold. The error of this atan2 implementation remains
below 25 ULP in most of its domain except for a neighborhood of y=0
where it reaches a maximum error of about 180 ULP.
- It emits a bunch of instructions including no less than three
if-else branches per scalar component that don't seem to get
optimized out later on. This implementation uses about 13% less
instructions on Intel SKL hardware and doesn't emit any control
flow instructions.
v2: Fix up argument scaling to take into account the range and
precision of exotic FP24 hardware. Flip coordinate system for
arguments along the vertical line as if they were on the left
half-plane in order to avoid division by zero which may give
unspecified results on non-GLSL 4.1-capable hardware. Sprinkle in
some more comments.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This does point at the front-end emitting silly code that could have
been optimized out, but the current fsign implementation would emit
bogus IR if abs was set for the argument (because it would apply the
abs modifier on an unsigned integer type), and we shouldn't rely on
the upper layer's optimization passes for correctness.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Will avoid a regression in a future commit that introduces some
additional rcp operations. According to the GLSL 4.10 specification:
"Dividing by 0 results in the appropriately signed IEEE Inf."
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
This will be used internally by the GLSL front-end in order to
implement some built-in functions. Plumb it through MESA IR for
back-ends that rely on this translation pass.
v2: Add comment.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
This fixes rendering of full-screen quads (and other screen-filling
geometry, e.g. ioquake3 walls up-close) on gc3000. It should be a no-op
on other hardware.
- It looks like SE_CLIP registers were not set at all.
I'm amazed that rendering worked without them. Emit them to
avoid issues on gc3000.
- Define constants
ETNA_SE_SCISSOR_MARGIN_RIGHT (0x1119)
ETNA_SE_SCISSOR_MARGIN_BOTTOM (0x1111)
ETNA_SE_CLIP_MARGIN_RIGHT (0xffff)
ETNA_SE_CLIP_MARGIN_BOTTOM (0xffff)
These demarcate the margin (fixp16) between the computed sizes and the
value sent to the chip. I have set these to the numbers used by the
Vivante driver for gc2000. I am not sure whether any old hardware was
relying on the old numbers, or whether those were just a guess. But if
so, these need to be moved to the _specs structure.
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Acked-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Shaders using sin/cos instructions were not working on GC3000.
The reason for this turns out to be that these chips implement sin/cos
in a different way (but using the same opcodes):
- Need their input scaled by 1/pi instead of 2/pi.
- Output an x and y component, which need to be multiplied to
get the result.
- tex_amode needs to be set to 1.
Add a new bit to the compiler specs and generate these instructions
as necessary.
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Acked-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Commit 2852efcda4 moved the location of
the depth input attachment surface state from the render pass to the
image view, but failed to update the surface state location used when
emitting the binding table. Fix this by loading the surface state from
the correct location.
Fixes:
dEQP-VK.renderpass.formats.d16_unorm.input.*
dEQP-VK.renderpass.formats.d24_unorm_s8_uint.input.*
dEQP-VK.renderpass.formats.d32_sfloat.input.*
dEQP-VK.renderpass.formats.x8_d24_unorm_pack32.input.*
dEQP-VK.renderpass.attachment_allocation.input_output.93
dEQP-VK.renderpass.attachment_allocation.input_output.92
dEQP-VK.renderpass.attachment_allocation.input_output.82
dEQP-VK.renderpass.attachment_allocation.input_output.46
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
The `end+1` skips the ']', whereas the `strlen+1` includes the final
'\0' in the move to terminate the string.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Exposing rb swapped (or other swizzled) formats for rendering would
involve swizzing in the pixel shader. This is not the case at the
moment, so reject requests for creating such surfaces.
(GPUs that need an extra resolve step anyway due to multiple pixel
pipes, such as gc2000, might also do this swap in the resolve operation.
But this would be tricky to keep track of)
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Acked-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Use of unsigned loop control variable with '>= 0' would lead
to infinite loop.
Reported by clang:
etnaviv_compiler.c:1024:39: warning: comparison of unsigned expression
>= 0 is always true [-Wtautological-compare]
for (unsigned sp = c->frame_sp; sp >= 0; sp--)
~~ ^ ~
v2: Simply use the same datatype as c->frame_sp is using.
CC: <mesa-stable@lists.freedesktop.org>
Reported-by: Rhys Kidd <rhyskidd@gmail.com>
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Rhys Kidd <rhyskidd@gmail.com>
There are some corner cases where you end up with an esgs ring, but no
gsvs ring, test for both before dereferencing.
Fixes:
dEQP-VK.geometry.emit.points_emit_0_end_0
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Fixes:
dEQP-VK.geometry.layered.1d_array.fragment_layer
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This emits the compiled geometry shader and other state registers.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This uses the scratch infrastructure to handle the esgs
and gsvs rings.
(this replaces the old code that did this with patching).
v2: fix correct ring sizes, reset sizes (Bas)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds gs copy shader support to the pipeline cache, and few
geometry related changes.
v2: rebase for spill changes.
v2.1: fix incorrect pipeline destruction.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This handles geometry shader inputs written by the vertex (es) shader
to the esgs ring.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Lionel Landwerlin
Emil Velikov
Michel Dänzer
Tapani Pälli
Ilia Mirkin
Jason Ekstrand
Tom Stellard
Bas Nieuwenhuizen
Rob Herring
Francisco Jerez
Wladimir J. van der Laan
Nanley Chery
Bartosz Tomczyk
Christian Gmeiner
Dave Airlie