Specifically, constant fold intrinsics from ARB_shader_group_vote, but I
suspect it'll be useful for other things in the future.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These are intrinsics rather than opcodes, because they operate across
channels.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
If the source is an indirect register, there is ralloc'd data. Copying
with a direct assignment will copy the pointer, but the data will still
belong to the old instruction's memory context. Since we're lowering
and throwing away instructions, that could free the data by mistake.
Instead, use nir_src_copy, which properly handles this.
This is admittedly not a common case, so I think the bug is real,
but unlikely to be hit.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
While it produces functioning code the pass creates worse code
for arrays of arrays. See the comment added in this patch for more
detail.
V2: skip splitting of AoA of matrices too.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Previously we only incremented the guide for a single
dimension/wildcard.
V2: rework logic to avoid code duplication
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: mesa-stable@lists.freedesktop.org
This can happen if, for instance, you have an array of structs and there
are both direct and wildcard references to the same struct and some
members only have direct or only have indirect.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Cc: mesa-stable@lists.freedesktop.org
Generated C files try to include spirv_info.h. For in-tree builds,
the header is in the same directory, so it just works. For out-of-tree
builds, we need to look for it in srcdir rather than builddir.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101831
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Now that vtn_type has piles of unions, we should assert sanity before
setting fields that may stomp others.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The old table based spirv_*_to_string functions would return NULL for
any values "inside" the table that didn't have entries. The tables also
needed to be updated by hand each time a new spirv.h was imported.
Generate the file instead.
v2: Make this script work more like src/mesa/main/format_fallback.py.
Suggested by Jason. Remove SCons supports. Suggested by Jason and
Emil. Put all the build work in Makefile.nir.am in lieu of adding a new
Makefile.spirv.am. Suggested by Emil. Add support for Android builds
based on code provided by Emil.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
The compact flag doesn't make sense on local variables, since the
packing on them is up to the driver. This fixes nir_validate assertions
in some cases, particularly when lower_io_to_temporaries is used on
per-vertex inputs/outputs.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
While normally we give variables whose name field is NULL a temporary
name when called from nir_print_shader(), when we were calling from
nir_print_instr() we never bothered, meaning that we just segfaulted
when trying to print out instructions with such a variable. Since
nir_print_instr() is meant to be called while debugging, we don't need
to bother too much about giving a consistent name, but we don't want to
crash in the middle of debugging.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
We were hitting the
unreachable("Invalid image opcode")
near the end of vtn_handle_image when parsing the
SpvOpAtomicCompareExchange opcode.
v2: Add stable CC.
v3: Ignore SpvOpAtomicCompareExchangeWeak. It requires the Kernel
capability which is not exposed in Vulkan, and spirv_to_nir is not used
for OpenCL which does support it.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
CC: <mesa-stable@lists.freedesktop.org>
Fixes a bunch of gl_BackColor interpolation tests that had explicit
interpolation specified on the fragment shader gl_Color.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>
It should have been removed after 00c47e111c.
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: Connor Abbott <cwabbott0@gmail.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
_mesa_glsl_has_builtin_function is used to determine whether any variant
of a builtin are available, for the purpose of enforcing the GLSL ES
3.00+ rule that overloads or overrides of builtins are disallowed.
However the builtin_builder contains information on all builtins,
irrespective of parse state, or versions, or extension enablement. As a
result we would say that a builtin existed even if it was not actually
available.
To resolve this, first check if at least one signature is available for
a builtin before returning true.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101666
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Now that we have proper pointer types, we can be more sensible about the
way we set up function arguments and deal with the two cases of pointer
vs. SSA parameters distinctly.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
We're going to want the full vtn_type available to us anyway at which
point glsl_type isn't really buying us anything.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This adds a vtn concept of base_type as well as a couple of other
fields. This lets us be a tiny bit more efficient in some cases but,
more importantly, it will eventually let us express things the GLSL type
system can't.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Now that we have a pointer wrapper class, we can create offsets for UBOs
and SSBOs up-front instead of waiting until we have the full access
chain. For push constants, we still use the old mechanism because it
provides us with some nice range information.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Everyone now calls it with stop_at_matrix = false. Since we're now
always walking all the way to the end of the access chain, the type
returned is just the same as ptr->type;
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Instead of handling all of the complexity at the end, we choose to
decorate types a bit more cleverly. When we have a row-major matrix
type, we give it the stride of a single vector and give it's array
element type (which represents a column) the actual matrix stride.
Previously, we were using stop_at_matrix and handling everything from
matrix on down as special cases but now we walk the access chain all the
way to the end and then load. Even though this looks like it may lead
to a significant functional change, it doesn't. The reason why we
needed to do stop_at_matrix before was to handle row-major properly
since the offsets and strides would be all out-of-order. Now that row
major matrix types have the small stride on the matrix and the large
stride on the vector, offsetting to a single column of a row-major
matrix works fine. The load/store code simply picks up on the fact that
the stride isn't the type size and does multiple loads. The generated
code from these methods should be the same.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
The vtn_pointer structure provides a bit better abstraction than passing
access chains around directly. For one thing, if the pointer just
points to a variable, we don't need the access chain at all. Also,
pointers know what their dereferenced type is so we can avoid passing
the type in a bunch of places. Finally, pointers can, in theory, be
extended to the case where you don't actually know what variable is
being referenced.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
We're about to add a vtn_pointer data structure and this will prevent
some rename churn in the next commit.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
We were originally handling them together because I was rather unclear
on the distinction. However, keeping them combined keeps the confusion.
Split them up so that it's more clear from the code how we expect the
two storage classes to be used.
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Previously, we were using the type of the variable which is incorrect.
Cc: "17.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
It's closing a "{" at the begining of a switch case.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This is convenient for backends that support both Vulkan and OpenGL while
lowering samplers to derefs with nir_lower_samplers_as_deref.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Undefined data will eventually trigger a valgrind error while computing
its CRC32 while writing it into the disk cache, but at that point, it is
basically impossible to track down where the undefined data came from.
With this change, finding the origin of undefined data becomes easy.
v2: remove duplicate VALGRIND_CFLAGS (Emil)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Otherwise, the padding bits remain undefined, which leads to valgrind
errors when storing the gl_shader_variable in the disk cache.
v2: use rzalloc instead of an explicit padding member variable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Save some passes over the IR.
v2: redesign to make the users of find_assignments more readable
v3:
- fix missing !
- add some comments and make the num_found check more explicit (Timothy)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> (v1)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Before, we were just implementing it with a move, which is incorrect
when the source and destination have different bitsizes. To implement
it properly, we need to use the 64-bit pack/unpack opcodes. Since
glslang uses OpBitcast to implement packInt2x32 and unpackInt2x32, this
should fix them on anv (and radv once we enable the int64 capability).
v2: make supporting non-32/64 bit easier (Jason)
v3: add another assert (Jason)
Fixes: b3135c3c ("anv: Advertise shaderInt64 on Broadwell and above")
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
