On GFX9 whether the buffer size is interpreted as elements or bytes
depends on whether IDXEN is enabled in the instruction. If the index
is a constant zero, LLVM optimizes IDXEN to 0.
Now the size in elements is interpreted in bytes which of course
results in out of bounds accesses.
The correct fix is most likely to disable the LLVM optimization,
but we need something to work with LLVM <= 6.0.
radeonsi does the max between stride and element count on the CPU
but that results in the size intrinsics returning the wrong size
for the buffer. This would cause CTS errors for radv.
v2: Also include the store changes.
Fixes: e38685cc62 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
I assume this was implemented in a previous version of that commit, but
was removed in the version that actually landed.
Fixes: 8430af5ebe "Add support for swrast to the DRM EGL platform"
Cc: Giovanni Campagna <gcampagna@src.gnome.org>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
The power-of-two padded size that gets minified is based on level 1's
dimensions, not level 0's, which starts to differ at a width of 9.
Fixes all failures on texelFetch fs sampler2D 1x1x1-64x64x1
bo->align is always 0; there's no need to waste 8 bytes storing it.
Thanks to C99 initializers zeroing fields, we can completely drop the
only read of the field altogether.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Buffers are always page aligned on 965+ hardware; I believe this extra
parameter is a vestige from the Gen2-3 era.
All callers pass 0, and in fact we assert that the alignment is 0 unless
BO_ALLOC_BUSY is set (for some reason). We can just drop the parameter
and set the value to 0 explicitly.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
intel_miptree_create_for_bo does not actually allocate a BO, so
specifying allocation flags accomplishes nothing and is confusing.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This is just zero - passing nothing already gives us a post-sync
operation of "nothing".
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
I have no idea why but having dest_components == -1 was causing a memory
leak somewhere. Without this, you can't get through a full shader-db
run without running out of memory.
Reviewed-by: Rob Clark <robdclark@gmail.com>
This was causing us to walk dest_components times over a thing with no
destination. This happened to work because all of the image intrinsics
without a destination also happened to have dest_components == 0. We
shouldn't be reading dest_components if has_dest == false.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
There were two problems, both of which are fixed now:
- The indirect address was not being shifted by 4
- The indirect address was being placed as an argument in the offset case
This fixes some of the new interpolateAt* piglits which now test for
these situations.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
When an if nesting inside anouther if is optimised away we can
end up with a loop terminator and following block that looks like
this:
if ssa_596 {
block block_5:
/* preds: block_4 */
vec1 32 ssa_601 = load_const (0xffffffff /* -nan */)
break
/* succs: block_8 */
} else {
block block_6:
/* preds: block_4 */
/* succs: block_7 */
}
block block_7:
/* preds: block_6 */
vec1 32 ssa_602 = phi block_6: ssa_552
vec1 32 ssa_603 = phi block_6: ssa_553
vec1 32 ssa_604 = iadd ssa_551, ssa_66
The problem is the phis. Loop unrolling expects the last block in
the loop to be empty once we splice the instructions in the last
block into the continue branch. The problem is we cant move phis
so here we lower the phis to regs when preparing the loop for
unrolling. As it could be possible to have multiple additional
blocks/ifs following the terminator we just convert all phis at
the top level of the loop body for simplicity.
We also add some comments to loop_prepare_for_unroll() while we
are here.
Fixes: 51daccb289 "nir: add a loop unrolling pass"
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105670
The location was only being incremented the first time we processed a
location. This meant we would incorrectly skip some elements of
an array if the first element was packed and proccessed previously
but other elements were not.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
We need to wait until after the writemask is widened before we
adjust it for component packing.
Together with the previous patch this fixes a number of
arb_enhanced_layouts component layout piglit tests.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This fixes tcs/tes varying arrays where we dont lower indirects and
therefore don't split arrays. Here we also fix useagemask for dual
slot doubles.
Fixes a number of arb_tessellation_shader piglit tests.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
How many times did I look at this table without noticing the missing 'G'
in the texture column?
Fixes KHR-GLES3.copy_tex_image_conversions.required.* on 7268.
Apparently it is not happy about things like: .foo = {}
So skip over initializers for empty lists.
Fixes: 76dfed8ae2
Reported-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>
I threatened to do this a long time ago.. I probably *should* have done
it a long time ago when there where many fewer intrinsics. But the
system of macro/#include magic for dealing with intrinsics is a bit
annoying, and python has the nice property of optional fxn params,
making it possible to define new intrinsics while ignoring parameters
that are not applicable (and naming optional params). And not having to
specify various array lengths explicitly is nice too.
I think the end result makes it easier to add new intrinsics.
v2: couple small fixes found with a test program to compare the old and
new tables
v3: misc comments, don't rely on capture=true for meson.build, get rid
of system_values table to avoid return value of intrinsic() and
*mostly* remove side-effects, add autotools build support
v4: scons build
Signed-off-by: Rob Clark <robdclark@gmail.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
EXT_color_buffer_float spec states:
"An INVALID_OPERATION error is generated ... if the color buffer is
a floating-point format and type is not FLOAT, HALF FLOAT, or
UNSIGNED_INT_10F_11F_11F_REV."
This means that GL_HALF_FLOAT type should be supported when color
buffer has floating-point format.
Fixes Android CTS test android.view.cts.PixelCopyTest.
v2: remove comments of EXT_color_buffer_half_float as
EXT_color_buffer_float can use type GL_HALF_FLOAT
Signed-off-by: Lin Johnson <johnson.lin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This is the actual hardware layout, and we were only swizzling R/B back
around in texturing. Fixes part of
KHR-GLES3.copy_tex_image_conversions.required.cubemap_negx_cubemap_negx in
simulation.
Once we've disabled EZ for some draws, we need to not use EZ on future
draws. Implementing that made implementing the GT/GE direction trivial.
Fixes KHR-GLES3.shaders.fragdepth.compare.no_write on V3D 4.1 simulation.
On 3.x, we just don't flag the primitive as needing TF, but those
primitive bits are now allocated to the new primitive types. Now we need
to actually update the enable flag at draw time.
The next job from this client will turn it back on unless TF gets
disabled, but we don't want the state to leak from this client to another
(which causes GPU hangs).
I had this note to myself, and it turns out that a lot of CTS tests use
XFB with points to get data out without using a fragment shader. Keep
track of two sets of precomputed TF specs (point size in VPM prologue or
not), and switch between them when we enable/disable point size.
Daniel Schürmann