..and assert the other sources are null. The one place this might fail in the
future is for real FMA, but we don't support that for GL.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16798>
These do not convey any additional information, and fail to account for
shrinking. In particular, a 64-bit writemask with .keephi would fail to
disassemble and instead trip the assertion, since that would be the ZW
components. Just delete the broken code.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16798>
Otherwise, we can get vec3 with u2u32 with 64-bit sources which we need lowered.
Since our current approach is "scalarize all 64-bit ops", we need to check for
conversions too.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16798>
The callback allows to request different vectorization factors
per instruction depending on e.g. bitsize or opcode.
This patch also removes using the vectorize_vec2_16bit option
from nir_opt_vectorize().
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13080>
NIR already has the necessary lowering, and the GLSL lowering violates
GLSL IR validation rules. Once quadop lowering was turned off, the IR
validation at the end of the compile path on DEBUG builds caught the
problem.
In order to move the lowering to NIR, though, we need to make sure that
drivers supporting these functions actually have the lowering flag set.
xfails added for t860, where apparently this tickles a variety of existing
64-bit bugs in the backend.
Fixes: #6461
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Mykhailo Skorokhodov <mykhailo.skorokhodov@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16437>
This is set to true for all drivers that have a GLSL level
of support lower than 4.00. This matches the rule for setting the
GLSL IR option EmitNoIndirectSampler.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16543>
Vulkan doesn't need nearly as many system values and would like to bake
its layout up-front instead of having it provided by the back-end
compiler.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16276>
In 3559efb9bf ("panfrost: Allow passing an explicit UBO index for the
sysval UBO"), an explicit UBO index was added and it was implicitly
assumed that it would be > num_ubos. This was convenient because it
meant 0, the default for designated initializers, implicitly meant
compiler-assigned. However, we're about to move the sysval UBO to 0
which breaks this assumption. Also, we don't want the back-end
compiler to even look at num_ubos since it's meaningless in Vulkan.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16276>
If two instructions in a single bundle both write to a spilt
destination, then we need to reuse the fill and spill instructions,
otherwise the value will be overwritten.
This and the rest of this set of Midgard bug fixes were found from a
vertex shader in Firefox WebRender that is used when a video is
clipped, for example by setting the border-radius CSS property.
CC: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16382>
Check the bytemask against 0xFFFF rather than 0xF so that the fill is
skipped for a .xyzw write rather than a .x write.
Set the mask on the store to 0xF when doing a read so that all
components are written back.
Fixes: 31d26ebf1b ("pan/mdg: Fill from TLS before spilling non-SSA nodes")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16382>
If a value is written in a vector CSEL but then written again by other
instructions, it still needs full alignment, so set min_alignment
using MAX2 to avoid ever reducing it.
Fixes: 1798f6bfc3 ("pan/midgard: Fix masks/alignment for 64-bit loads")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16382>
This controls the whole lowering of "make tex ops with implicit
derivatives on non-implicit-derivative stages be tex ops with an explicit
lod of 0 instead", but it's really hard to describe that in a git commit
summary.
All existing callers get it added except:
- nir_to_tgsi which didn't want it.
- nouveau, which didn't want it (fixes regressions in shadowcube and
shadow2darray with NIR, since the shading languages don't expose txl of
those sampler types and thus it's not supported in HW)
- optional lowering passes in mesa/st (lower_rect, YUV lowering, etc)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16156>
The GLSL lowering of half float packing involves software conversion
to half-float; instead, use the lowering in NIR.
Both Midgard and Bifrost are already set to lower the instructions to
bit operations, but change mdg_should_scalarize so that the lowerable
split variants of the pack/unpack instructions are generated.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16175>
This is to match other NIR terminology.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15103>
BITSET_MASK returns ~0 when given an input of zero, when we need it to
return 0 instead.
Fixes shaders with only sysvals but no UBOs when push constants are
disabled.
This breaks when 31 or 32 UBOs are used, but PAN_MAX_CONST_BUFFERS is
currently set to 16.
Fixes: c246af0dd8 ("panfrost: Only upload UBOs when needed")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15250>
There are up to 4 instructions in the latter stage (if a branch is included),
not 3. Bump the limit to fix memory corruption.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reported-by: Icecream95 <ixn@disroot.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15147>
We always blit from a particular level, so it's a waste to compute the LOD.
This corresponds to a simple texture instruction with implement 0 LOD, which is
the optimal texturing path on Bifrost -- it maps to TEXS_2D but does not require
helper invocations.
Functional change on Bifrost: Blit shaders no longer set .computed_lod or
shader_contains_barrier.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15123>
cont -> skip, last -> kill, and fix the special case handling. It's just an
enum. Makes the disassembly easier to read and closer to Bifrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15123>
To help identify problems across the compiler, print more forms of the shader
with MIDGARD_MESA_DEBUG=shaders. Roughly matches the Bifrost compiler.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14888>
Commit 38800b38 changed nir_opcodes.py, but that doesn't seem to have
triggered nir_opt_algebraic.py. The change in 75ef5991 depends on
opt_algebraic lowering 16-bit versions of slt, but if opt_algebraic is
not rebuilt, this may not happen. This resulted in some people seeing
assertion failures in, for example,
dEQP-VK.spirv_assembly.instruction.compute.float16.arithmetic_3.step,
due to the backend seeing nir_op_slt that it didn't know how to handle.
v2: Add nir_opcodes.py to nir_algebraic_py so that all the per-driver
algebraic passes pick up the dependency too. Rename it to
nir_algebraic_depends. Suggested by Emma.
Closes: #6047
Fixes: d1992255bb ("meson: Add build Intel "anv" vulkan driver")
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15050>
Compute shaders might do vec3(Xbits) loads which are translated
to LD.<next_pow2(3 * X)> by the midgard compiler. This might cause
out-of-bound accesses potentially leading to pagefaults if the
access is at the end of a BO. One solution to avoid that (maybe not
the best) is to split non-log2 loads to make sure we only read what's
requested.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14885>
In that case, the mask is specified on 32bit lanes, so we need to shift
it if it's > 0x3. The expand modifier will take care of selecting the
right side of the 32bit vector.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14885>
Swizzling happens in 2 steps on Midgard:
1. Vector expansion/shuffling
2. Swizzling at the instruction-size granularity, but defined using
the source size. Those size are different if the source is expanded.
So, when we print 64 bit swizzles on an expanded source, we first need
to apply an offset if the high part of the 32bit vector was selected,
and then divide the result by 2 to account for vector expansion.
To sum-up, swizzling on midgard is complicated, and I'm not sure I got
it right, but it seems to print what I expect on the few compute
shaders using 64bit arithmetic I debugged.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14885>
Even though 8-bit ALUs are not supported, we can have [un]pack_32_4x8
instructions which translate to IMOVs, and those operate on 8-bit
vectors. The problem is, the swizzling granularity is 16 bit, which
means we don't support
MOV.i8 R0.xyzw, TMP0.xxxx, R1.zyxw
and the compiler doesn't even complain, it just applies 8 bit
swizzling directly, which obviously doesn't work.
This is probably not the right way to fix that, but I thought I'd
raised the issue with a hack to fix, so we can get the discussion
started.
(Found while debugging FB store lowering on Midgard).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14885>
The register allocator assumes instructions are executed sequentially
and allows one instruction to overwrite a portion of a register written
by a previous instruction if this portion is never used. But scalar and
vector ALUs might be executed in parallel if they are part of the same
bundle, and when such instructions write to the same portion of the
register file, the result is undefined.
Let's add intra-bundle interferences to avoid this situation.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14885>