From @jekstrand's nir-1-bit-bool branch, with improved ior/inot lowering.
ior: fmax instead of fadd allows removing the fsat.
inot: seq(x, 0) can be better than fsub(1, x). On a2xx, it works better
with the scalar instruction set.
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Replace calls to create hash tables and sets that use
_mesa_hash_pointer/_mesa_key_pointer_equal with the helpers
_mesa_pointer_hash_table_create() and _mesa_pointer_set_create().
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Eric Engestrom <eric@engestrom.ch>
Function's out variable could be an array dereferenced by an array:
func(v[w[i]]);
or something more complicated.
Copy index in any case.
Fixes: 76c27e47b9 ("glsl: Copy function out to temp if we don't directly ref a variable")
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Instead of emitting all of the conditions for the cases of a switch
statement up-front, emit them on-the-fly as we emit the code for each
case. The original justification for this was that we were going to
have to build a default case anyway which would need them all. However,
we can just trust CSE to clean up the mess in that case. Emitting each
condition right before the if statement that uses it reduces register
pressure and, in one customer benchmark, reduces spilling and improves
performance by about 2x.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Even though no one's been brave enough to ever use this pass, I like to
keep it functionally working.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Instead of applying the workaround universally, detect semi-old GLSLang
via the generator ID and only enable the workaround on old GLSLang.
This isn't nearly as precise as one would like it to be because the
first GLSLang generator id version bump was on October 7, 2017 which is
about 1.5 years after the bug was fixed. However, it at least lets us
disable it for non-GLSLang and for more modern versions.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
A long time in a galaxy far far away, there was a GLSLang bug with how
it handled samplers passed in as function parameters. (The bug can be
found here: https://github.com/KhronosGroup/glslang/issues/179.)
Unfortunately, that version was shipped in several apps and has been
causing heartburn for our SPIR-V parser ever since.
Recent changes to NIR uncovered a moderately old bug in how we work
around this issue. In particular, we ended up with a deref_cast from
uniform to local which is not a no-op cast so nir_opt_deref wasn't
getting rid of the cast. The only reason why it worked before was
because someone just happened to call nir_fixup_deref_modes which
"fixed" the cast (that shouldn't be happening) and then a later round of
copy-prop would get rid of it. The fact that the deref_cast survived
that long without causing trouble for other parts of NIR is a bit
surprising.
Just whacking the mode of the pointer seems to fix it fairly
unobtrusively. Currently, only apps with this bug will have a local
variable containing an image or sampler.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109304
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
With the new handling of bool types, the conversion to float in glsl_to_nir
should not apply to bool types anymore.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
out_type in the default cast case is always GLSL_TYPE_FLOAT, so we get a
mov otherwise.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
All alu instructions emitted with native_integers=false expect float
(or bool in some cases) constants, so this change is necessary.
This will cause changes with some intrinsics which had integer sources,
such as nir_intrinsic_load_uniform. Apparently it might cause issues with
some opt passes, but perhaps those don't apply in OpenGL ES 2.0 cases?
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
GL_ARB_gl_spirv does not provide a sampler deref for e.g. texelFetch(),
so we can't assume that both are present and identical. Simply lower
each if it is present.
Fixes regressions in GL_ARB_gl_spirv tests since I switched everyone to
using this pass. Thanks to Alejandro Piñeiro for catching these.
Fixes: f003859f97 nir: Make gl_nir_lower_samplers use gl_nir_lower_samplers_as_deref
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Alejandro Piñeiro <apinheiro@igalia.com>
In all GLSL ES versions output variables in fragment shader are allowed
to be invariant.
From Section 4.6.1 ("The Invariant Qualifier") GLSL ES 1.00 spec:
"Only the following variables may be declared as invariant:
...
- Built-in special variables output from the fragment shader."
From Section 4.6.1 ("The Invariant Qualifier") GLSL ES 3.00 spec:
"Only variables output from a shader can be candidates for invariance."
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107842
They have glsl_sampler_dim enum values of 8 and 9 which don't work when
you & them with 0x7. Fortunately, we have plenty of bits.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 393b59e077
('nir: Rework nir_lower_constant_initializers() to handle functions')
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The check for location aliasing was always asuming output variables
but this validation is also called for input variables.
Fixes: e2abb75b0e ("glsl/linker: validate explicit locations for SSO programs")
Cc: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
NIR metadata validation verifies that the debug bit was unset (by a call
to nir_metadata_preserve) if a NIR optimization pass made progress on
the shader. With the expectation that the NIR shader consists of only a
single main function, it has been safe to call nir_metadata_preserve()
iff progress was made.
However, most optimization passes calculate progress per-function and
then return the union of those calculations. In the case that an
optimization pass makes progress only on a subset of the functions in
the shader metadata validation will detect the debug bit is still set on
any unchanged functions resulting in a failed assertion.
This patch offers a quick solution (short of a larger scale refactoring
which I do not wish to undertake as part of this series) that simply
unsets the debug bit on unchanged functions.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Will be used to communicate that a shader uses 64-bit operations to the
concerned lowering passes.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We're going to have multiple functions, so nir_shader_get_entrypoint()
needs to do something a little smarter.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Previously it assumed that only a single function (the entrypoint)
existed and attempted to lower constant initializers of shader outputs
for each function, for instance.
v2: use mix and findMSB to optimise.
v3: [Sagar] Fix zFrac0 == 0u case in __normalizeRoundAndPackFloat64
Signed-off-by: Elie Tournier <elie.tournier@collabora.com>
The following patches will add implementations of various
double-precision operations to this file.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Will be used to convert the .glsl source file containing software fp64
routines to a .h file that can be included while building the compiler.
This commit contains two squashed together: the first from Ian adding
the utility (with the existing title), and the second from Dylan making
the code both python2 and python3 compatible.
This is somewhat modeled after the xxd utility that comes with Vim.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
xxd.py: Make python2 and 3 compatible
This makes use of unicode_literals, so that undecorated strings are
considered text (python2 unicode, python3 str) and not bytes in python2
and text in python3. It makes use of io.open, which provides python2
with python3's open behavior (it's an alias in python3), in particular
support for the 't' and 'b' option. Finally, it decorates all of the
string literals with the 'b' prefix, so that python interprets them as
bytes.
I've removed the stdin and stdout options, as python2 always requires
these to be bytes, but python3 always treats them as text (there is a
way to get at the underlying bytes buffer, but that's even more
complexity), and makes the input files required arguments.
In the meson we use the '@INPUT@' shorthand instead of listing each
input, as meson will expand that to [prog_python, '@INPUT0@', @INPUT1@,
..., @OUTPUT@, ...]
Otherwise we can end up with IR that looks like this:
(
(declare (temporary ) vec4 f@8)
(assign (xyzw) (var_ref f@8) (var_ref f) )
(call f16 ((swiz y (var_ref f@8) )))
(assign (xyzw) (var_ref f) (var_ref f@8) )
))
When we really need:
(declare (temporary ) float inout_tmp)
(assign (x) (var_ref inout_tmp) (swiz y (var_ref f) ))
(call f16 ((var_ref inout_tmp) ))
(assign (y) (var_ref f) (swiz y (swiz xxxx (var_ref inout_tmp) )))
(declare (temporary ) void void_var)
The GLSL IR function inlining code seemed to produce correct code
even without this but we need the correct IR for GLSL IR -> NIR to
be able to understand whats going on.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Based on a patch from Tim Arceri, but I had to substantially rewrite it
as a result of the NIR derefs rework.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The binding is checked against the limits later in the function, so we
need to make sure we don't overflow before the check here.
Fixes this valgrind warning (and sometimes segfault):
==1460== Invalid write of size 4
==1460== at 0x74C98DD: ast_declarator_list::hir(exec_list*, _mesa_glsl_parse_state*) (ast_to_hir.cpp:4943)
==1460== by 0x74C054F: _mesa_ast_to_hir(exec_list*, _mesa_glsl_parse_state*) (ast_to_hir.cpp:159)
==1460== by 0x7435C12: _mesa_glsl_compile_shader (glsl_parser_extras.cpp:2130)
in
dEQP-GLES31.functional.debug.negative_coverage.get_error.compute.
exceed_atomic_counters_limit
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
I think this was copy-and-paste mistake -- nir_opt_large_constants was
passing in glsl_get_natural_size_align_bytes() given brw_nir.c's arguments
to the opt pass.
I wanted to reuse this function for handling constant offsets of arrays of
images in V3D.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Always match TCS outputs since they are shared by all invocations
within the patch and should not be converted to local variables.
This is one of the issues found in Downward.
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104297
V3D returns the texels in a different order in the resulting vec4 from
what GLSL wants, so we need to put in a swizzle. Fixes
dEQP-GLES31.functional.texture.gather.basic.2d.rgba8.base_level.level_1
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Initialize the variable with NULL. Fixes the following
In file included from ../src/compiler/nir/nir_lower_io.c:34:
../src/compiler/nir/nir_lower_io.c: In function ‘nir_lower_explicit_io’:
../src/compiler/nir/nir.h:668:11: warning: ‘addr’ may be used uninitialized in this function [-Wmaybe-uninitialized]
return src;
^~~
../src/compiler/nir/nir_lower_io.c:735:17: note: ‘addr’ was declared here
nir_ssa_def *addr;
^~~~
v2: Avoid using a 'default' case so we get help from the compiler when
new deref types are added. (Lionel)
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
the naming is a bit confusing no matter how you look at it. Within SPIR-V
"global" memory is memory accessible from all threads. glsl "global" memory
normally refers to shader thread private memory declared at global scope. As
we already use "shared" for memory shared across all thrads of a work group
the solution where everybody could be happy with is to rename "global" to
"private" and use "global" later for memory usually stored within system
accessible memory (be it VRAM or system RAM if keeping SVM in mind).
glsl "local" memory is memory only accessible within a function, while SPIR-V
"local" memory is memory accessible within the same workgroup.
v2: rename local to function as well
v3: rename vtn_variable_mode_local as well
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
For now, it's hidden behind a cap. Hopefully, we can eventually drop
that along with all the manual offset code in spirv_to_nir.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
The choice of whether or not we should use block_load/store isn't a
choice between external and not so much as a choice between deref
instructions and manually calculated offsets. In vtn_pointer_from_ssa,
we guard the index+offset case behind vtn_pointer_uses_ssa_offset and
then branch out from there.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Instead of baking in uvec2 for UBO and SSBO pointers and uint for push
constant and shared memory pointers, make it configurable.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Previously, we hard-coded the rule about workgroup variables and the
builder lower_workgroup_access_to_offsets flag. Instead base it on the
handy helper we have for exactly this sort of thing.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Variable pointers being well-defined across the block boundary requires
a couple of very specific SPIR-V validation rules. Normally, we'd trust
the validator to catch these but since CTS tests have been found in the
wild which violate them, we'll carry our own checks.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This new pass is for lowering explicitly laid out memory coming in from
SPIR-V or a similar source. It's quite a bit more complicated than the
normal lower_io because we have to be able to handle matrices. The
way the stride information is stored for matrices is awkward and dealing
with row-major matrices is especially painful.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This commit adds a new num_components value for intrinsic sources of -1
which means that it consumes everything and the number of components
effectively isn't validated. This is useful for deref sources which
just take the result of the deref and we leave it up to the driver to
decide what that size should be.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We added this assert when first moving derefs over to instructions to
ensure that deref chains could go all the way back to the variables.
Now that we're going to start using derefs for things that we can do
variable pointers on such as UBOs and SSBOs, we need to be able to run
derefs through phi nodes, selects, and basically anything else.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We already detect any incomplete deref chains (where the deref is used
for something other than another deref or a load/store) and flag the
variable as used thanks to deref_used_for_not_store. All that's left to
do is to properly skip casts when cleaning up.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This pass is used when, for instance, we lazily change the mode of
variables rather than replacing the variable with a new one. Since we
only do this in cases where we know we have full deref chains, it's ok
to just skip them in fixup_deref_modes.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The code which constructs deref paths already gives you the path
starting at the nearest deref_cast or deref_var. All we need to do for
casts is handle the case where the start of the path isn't a deref_var.
For ptr_as_array derefs, we just bail if we have any after the
divergence point between the two derefs. We may be able to do better in
the future but this works for now.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
When handling casts, we can't blindly propagate the parent of a cast
into a ptr_as_array deref because doing so might loose the stride
information from the cast. Instead, before we can propagate into
ptr_as_array derefs, we need to check that the cast is a cast of an
array deref and that the stride matches. For other types of derefs, we
can continue to propagate casts as normal because they don't need the
stride. We also add an optimization which can combine a ptr_as_array
deref with it parent if it is also an array deref of some form.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
These correspond directly to SPIR-V's OpPtrAccessChain. As such, they
treat whatever their parent gives them as if it's the first element in
some array and dereferences that array. If the parent is, itself, an
array deref, then the two indices can just be added together to get the
final array deref. However, it can also be used in cases where what you
have is a dereference to some random vec2 value somewhere. In this
case, we require a cast before the ptr_as_array and use the ptr_stride
field in the cast to provide a stride for the ptr_as_array derefs.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We're going to want to do more deref optimizations going forward and
this gives us a central place to do them. Also, cast propagation will
get a bit more complicated with the addition of ptr_as_array derefs.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Instead of just storing the decorations in the vtn_type, propagate them
all the way through to the glsl_type. For array strides, this means we
need to handle them earlier so we break array stride handling into it's
own function and explicitly call it for both pointer and array types.
Due to type deduplication in the SPIR-V, we may have explicit layout
decorations on all sorts of types that don't actually want them. In
order to prevent these leaking into unfortunate places in NIR, we
explicitly strip them off before creating NIR variables and when casting
pointers to non-external memory.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
SPIR-V allows for matrix and array types to be decorated with explicit
byte stride decorations and matrix types to be decorated row- or
column-major. This commit adds support to glsl_type to encode this
information. Because this doesn't work nicely with std430 and std140
alignments, we add asserts to ensure that we don't use any of the std430
or std140 layout functions with explicitly laid out types.
In SPIR-V, the layout information for matrices is applied to the parent
struct member instead of to the matrix type itself. However, this is
gets rather clumsy when you're walking derefs trying to compute offsets
because, the moment you hit a matrix, you have to crawl back the deref
chain and find the struct. Instead, we take the same path here as we've
taken in spirv_to_nir and put the decorations on the matrix type itself.
This also subtly adds support for strided vector types. These don't
come up in SPIR-V directly but you can get one as the result of taking a
column from a row-major matrix or a row from a column-major matrix.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
This is C++ so we can just poke at the fields of glsl_type if we wish
and calling get_instance is way easier and more reliable than handling
each instance separately. While we're at it, we re-arrange the base
type labels to match the enum order and add 8-bit type support.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
It was added in bce6f99875 even though it's completely redundant with
glsl_array_type().
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Previously, NIR had a single nir_var_uniform mode used for atomic
counters, UBOs, samplers, images, and normal uniforms. This commit
splits this into nir_var_uniform and nir_var_ubo where nir_var_uniform
is still a bit of a catch-all but the nir_var_ubo is specific to UBOs.
While we're at it, we also rename shader_storage to ssbo to follow the
convention.
We need this so that we can distinguish between normal uniforms and UBO
access at the deref level without going all the way back variable and
seeing if it has an interface type.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
I have no idea how shader_storage made it into the list of banned
variable modes for stores but it clearly should be allowed. This only
doesn't cause us a problem today because we never actually use derefs on
shader_storage variables.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This doesn't currently change anything because array indices are
required to be 32 bits and all derefs are also 32 bits. However, we
will one day have 64-bit derefs for OpenCL.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We already had code in link_as_ssa to handle bit sizes; we just need to
use it. While we're at it we clean up link_as_ssa a bit and add an
explicit bit_size parameter in preparation for a day when we have derefs
that aren't 32 bit.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This simplifies our deref handling by emitting the actual NIR deref
instructions on-the-fly instead of of building up a deref chain and then
emitting them at the last moment. In order for this to work with the
parts of the compiler that assume they can chase deref chains, we have
to run nir_rematerialize_derefs_in_use_blocks_impl to put the derefs
back in the right places. Otherwise, in cases such as loop continues
where the SPIR-V blocks are not in the same order as the NIR blocks, we
may end up with a deref chain with a parent that does not dominate it's
child and nir_repair_ssa_impl will insert phis in the deref chain.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
The SPIR-V spec was recently updated to clarify that array indices are
treated as signed integers.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This crops up both in the actual SPIR-V VectorInsert/Extract opcodes as
well as various places where we deal with vector derefs.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
They can be handled exactly the same as arrays, we just need to handle
the base type correctly in the switches.
Fixes: a45b6fb452 "spirv: Pass SSA values through functions"
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109204
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
These days, we have two sampler lowering passes. The newer one,
gl_nir_lower_samplers_as_deref, is used by radeonsi. It rewrites
variables to drop structures out of sampler deref chains, to make
life simpler. It then sets var->data.binding for non-bindless
sampler and image variables based on the GL uniform storage's
opaque index values.
The older one converts sampler deref chains (nir_tex_src_texture_deref)
to a numerical offset (nir_tex_src_texture_offset). It also stores the
constant-valued portion of that number in tex->texture_index, making
life really simple for drivers that don't support indirects. It too
pokes at GL uniform storage's opaque index values.
Logically, we can do the first pass (simplify derefs, set bindings)
then the second (turn derefs to offsets, set texture_index). This
patch does exactly that, eliminating some redundancy (only one pass
has to poke at GL uniform storage), and gaining proper var->data.binding
values for drivers using the full lowering.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
We recurse to remove structures, and at each step, re-modify the
resulting type for our link in the deref chain. For arrays, the
result of recursion is the new underlying type - so we wrap it with
the array dimensionality again. For structs, we want to simply use
the new underlying type, skipping the struct altogether.
The correct way to do this is to do nothing at all. Previously, we
had reset type to next->type, which is the /old/ field type, not the
new field type we obtained by recursing. This undid our recursive work.
Fixes about 338 tests with nested structs, such as:
dEQP-GLES2.functional.uniform_api.value.initial.get_uniform.nested_structs_arrays.sampler2D_samplerCube_fragment
Note that currently only radeonsi uses this pass, and NIR support is
disabled there by default, so the breakage was likely not seen by most
people. The next commit uses this pass for more drivers, so this fix
prevents regressions from that change.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
I've been doing this in the nir-to-vir and nir-to-qir backends of v3d and
vc4, but nir could potentially do some useful stuff for us (like avoiding
unpack/repacks) if we give it the information.
v2: Skip lowering for txs/query_levels
v3: Fix a crash on old-style shadow
v4: Rename to tex_packing, use nir_format_unpack_sint/uint helpers, pack
the enum.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When copy_prop_vars also took care of dead write handling, intrin was
used as part of store_to_entry. Now it isn't, so this assignment
isn't used really used. Add a comment clarifying what happens to
intrin.
Fixes: 4dfa7adc10 "nir: Remove handling of dead writes from copy_prop_vars"
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Commit 27f1298b9d ("glsl/linker: validate attribute aliasing before optimizations")
forgot to complete the documentation.
Cc: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
After trying multiple times to merge if-statements with phis
between them I've come to the conclusion that it cannot be done
without regressions. The problem is for some shaders we end up
with a whole bunch of phis for the merged ifs resulting in
increased register pressure.
So this patch just merges ifs that have no phis between them.
This seems to be consistent with what LLVM does so for radeonsi
we only see a change (although its a large change) in a single
shader.
Shader-db results i965 (SKL):
total instructions in shared programs: 13098176 -> 13098152 (<.01%)
instructions in affected programs: 1326 -> 1302 (-1.81%)
helped: 4
HURT: 0
total cycles in shared programs: 332032989 -> 332037583 (<.01%)
cycles in affected programs: 60665 -> 65259 (7.57%)
helped: 0
HURT: 4
The cycles estimates reported by shader-db for i965 seem inaccurate
as the only difference in the final code is the removal of the
redundent condition evaluations and jumps.
Also the biggest code reduction (~7%) for radeonsi was in a tomb
raider tressfx shader but for some reason this does not get merged
for i965.
Shader-db results radeonsi (VEGA):
Totals from affected shaders:
SGPRS: 232 -> 232 (0.00 %)
VGPRS: 164 -> 164 (0.00 %)
Spilled SGPRs: 59 -> 59 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 14584 -> 13520 (-7.30 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 13 -> 13 (0.00 %)
Wait states: 0 -> 0 (0.00 %)
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Makes debugging easier when we care about the deref chain and not the
deref instruction itself. To make it take a const pointer, constify
some of the static functions in nir_print.c.
Reviewed-by: Eric Anholt <eric@anholt.net>
The 16-bit polynomial execution doesn't meet Khronos precision requirements.
Also, the half-float denorm range starts at 2^(-14) and with asin taking input
values in the range [0, 1], polynomial approximations can lead to flushing
relatively easy.
An alternative is to use the atan2 formula to compute asin, which is the
reference taken by Khronos to determine precision requirements, but that
ends up generating too many additional instructions when compared to the
polynomial approximation. Specifically, for the Intel case, doing this
adds +41 instructions to the program for each asin/acos call, which looks
like an undesirable trade off.
So for now we take the easy way out and fallback to using the 32-bit
polynomial approximation, which is better (faster) than the 16-bit atan2
implementation and gives us better precision that matches Khronos
requirements.
v2:
- Fallback to 32-bit using recursion (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fadd_imm and nir_fmul_imm helpers (Jason)
v3:
- since we need to define one for fsub use it for fdiv too (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- fix huge_val for 16-bit, it was mean't to be 2^14 not 10^14.
v3:
- rebase on top of new bool sized opcodes
- use nir_b2f helper
- use nir_fmul_imm helper
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fadd_imm and nir_fmul_imm helpers (Jason)
- rebased on top of new sized boolean opcodes
- use nir_b2f helper
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2:
- use nir_fmul_imm and nir_fadd_imm helpers (Jason)
v3:
- missed one case where we need to replace nir_imm_float
with nir_imm_floatN_t (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This just cleans things up a little and make things more safe for
derefs.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Eric Anholt <eric@anholt.net>
This will be reused by the following patch.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The following patches will add support for an additional
optimisation so this function will no longer just optimise varying
constants.
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The former expects to see SSA-only things, but the latter injects registers.
The assertions in the lowering where not seeing this because they asserted
on the bit_size values only, not on the is_ssa field, so add that assertion
too.
Fixes: 11dc130779 "nir: Add a bool to int32 lowering pass"
CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When copy propagation handles a store/copy, it iterates the current
copy entries to remove aliases, but keeps the "equal" entry (if
exists) to be updated.
The removal step may swap the entries around (to ensure there are no
holes), invalidating previous iteration pointers. The bug was saving
such pointer to use later. Change the code to first perform the
removals and then find the remaining right entry.
This was causing updates to be lost since they were being made to an
entry that was not part of the current copies.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108624
Fixes: b3c6146925 "nir: Copy propagation between blocks"
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
When updating a copy entry source value from a "non-SSA" (the data
come from a copy instruction) to a "SSA" (the data or parts of it come
from SSA values), it was possible to hold invalid data in ssa[0]
depending on the writemask. Because the union, ssa[0] could contain a
pointer to a nir_deref_instr left-over from previous non-SSA usage.
Change code to clean up the array before use to avoid invalid data
around.
Fixes: 62332d139c "nir: Add a local variable-based copy propagation pass"
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The quotation marks around 1.0 cause it to be treated as a string
instead of a floating point value. The generator then treats it as an
arbitrary variable replacement, so any iand involving a ('ineg', ('b2i',
a)) matches.
v2: Remove misleading comment about sized literals (suggested by
Timothy). Add assertion that the name of a varible is entierly
alphabetic (suggested by Jason).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Timothy Arceri <tarceri@itsqueeze.com> [v1]
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com> [v1]
Fixes: 6bcd2af086 ("nir/algebraic: Add some optimizations for D3D-style Booleans")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109075
Tested on gen9.
v2: Rename lower_txd_3d_surafaces flag to lower_txd_3d (Jason Ekstrand)
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Instead of going all the way back to the variable, just look at the
deref. The modes are guaranteed to be the same by nir_validate whenever
the variable can be found. This fixes clear_unused_for_modes for
derefs that don't have an accessible variable.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Instead of going all the way back to the variable, just look at the
deref. The modes are guaranteed to be the same by nir_validate whenever
the variable can be found. This fixes apply_barrier_for_modes for
derefs that don't have an accessible variable.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
This is instead of looking all the way back to the variable which may
not exist for all derefs. This makes this code properly ignore casts
with modes other than the mode[s] we care about (where casts aren't
allowed).
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
If we can't find the variable from the deref, just assume it isn't
invariant and continue on. This can happen if, for instance, we're
writing to a deref that points into an SSBO.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
"There's no point in walking the program if we're never going to
actually lower anything."
Except we might lower compacted local arrays. In that case, modes will
be 0, but there is still lowering to be done.
This reverts commit 7f75cf2a94.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109081
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Clayton Craft <clayton.a.craft@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
On some GPUs, especially older Intel GPUs, some math instructions are
very expensive. On those architectures, don't reduce flow control to a
csel if one of the branches contains one of these expensive math
instructions.
This prevents a bunch of cycle count regressions on pre-Gen6 platforms
with a later patch (intel/compiler: More peephole select for pre-Gen6).
v2: Remove stray #if block. Noticed by Thomas.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
That flow control may be trying to avoid invalid loads. On at least
some platforms, those loads can also be expensive.
No shader-db changes on any Intel platform (even with the later patch
"intel/compiler: More peephole select").
v2: Add a 'indirect_load_ok' flag to nir_opt_peephole_select. Suggested
by Rob. See also the big comment in src/intel/compiler/brw_nir.c.
v3: Use nir_deref_instr_has_indirect instead of deref_has_indirect (from
nir_lower_io_arrays_to_elements.c).
v4: Fix inverted condition in brw_nir.c. Noticed by Lionel.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
I botched some copy-and-paste and clamped to signed int max instead of
uint max. Fixes KHR-GL46.shader_image_load_store.multiple-uniforms on
skl.
Fixes: d3e046e76c ("nir: Pull some of intel's image load/store format
conversion to nir_format.h")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
nir_sweep already marks all metadata invalid, so it is safe to release
the memory here too.
mean soft fp64 using uint64: 1,342,759,331 => 1,010,670,475
gfxbench5 aztec ruins high 11: 63,555,571 => 61,889,811
deus ex mankind divided 148: 62,845,304 => 62,829,640
deus ex mankind divided 2890: 71,922,686 => 71,922,686
dirt showdown 676: 69,238,607 => 69,238,607
dolphin ubershaders 210: 77,822,072 => 77,822,072
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Found using pahole.
Changes in peak memory usage according to Valgrind massif:
mean soft fp64 using uint64: 1,343,991,403 => 1,342,759,331
gfxbench5 aztec ruins high 11: 63,619,971 => 63,555,571
deus ex mankind divided 148: 62,887,728 => 62,845,304
deus ex mankind divided 2890: 72,399,750 => 71,922,686
dirt showdown 676: 69,464,023 => 69,238,607
dolphin ubershaders 210: 78,359,728 => 77,822,072
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Replace the old array in each value with a hash table in each value.
Changes in peak memory usage according to Valgrind massif:
mean soft fp64 using uint64: 5,499,875,082 => 1,343,991,403
gfxbench5 aztec ruins high 11: 63,619,971 => 63,619,971
deus ex mankind divided 148: 62,887,728 => 62,887,728
deus ex mankind divided 2890: 72,402,222 => 72,399,750
dirt showdown 676: 74,466,431 => 69,464,023
dolphin ubershaders 210: 109,630,376 => 78,359,728
Run-time change for a full run on shader-db on my Haswell desktop (with
-march=native) is 1.22245% +/- 0.463879% (n=11). This is about +2.9
seconds on a 237 second run. The first time I sent this version of this
patch out, the run-time data was quite different. I had misconfigured
the script that ran the test, and none of the tests from higher GLSL
versions were run. These are generally more complex shaders, and they
are more affected by this change.
The previous version of this patch used a single hash table for the
whole phi builder. The mapping was from [value, block] -> def, so a
separate allocation was needed for each [value, block] tuple. There was
quite a bit of per-allocation overhead (due to ralloc), so the patch was
followed by a patch that added the use of the slab allocator. The
results of those two patches was not quite as good:
mean soft fp64 using uint64: 5,499,875,082 => 1,343,991,403
gfxbench5 aztec ruins high 11: 63,619,971 => 63,619,971
deus ex mankind divided 148: 62,887,728 => 62,887,728
deus ex mankind divided 2890: 72,402,222 => 72,402,222 *
dirt showdown 676: 74,466,431 => 72,443,591 *
dolphin ubershaders 210: 109,630,376 => 81,034,320 *
The * denote tests that are better now. In the tests that are the same
in both patches, the "after" peak memory usage was at a different
location. I did not check the local peaks.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
D3D Booleans use a 32-bit 0/-1 representation. Because this previously
matched NIR exactly, we didn't have to really optimize for it. Now that
we have 1-bit Booleans, we need some specific optimizations to chew
through the D3D12-style Booleans.
Shader-db results on Kaby Lake:
total instructions in shared programs: 15136811 -> 14967944 (-1.12%)
instructions in affected programs: 2457021 -> 2288154 (-6.87%)
helped: 8318
HURT: 10
total cycles in shared programs: 373544524 -> 359701825 (-3.71%)
cycles in affected programs: 151029683 -> 137186984 (-9.17%)
helped: 7749
HURT: 682
total loops in shared programs: 4431 -> 4399 (-0.72%)
loops in affected programs: 32 -> 0
helped: 21
HURT: 0
total spills in shared programs: 10290 -> 10051 (-2.32%)
spills in affected programs: 2532 -> 2293 (-9.44%)
helped: 18
HURT: 18
total fills in shared programs: 22203 -> 21732 (-2.12%)
fills in affected programs: 3319 -> 2848 (-14.19%)
helped: 18
HURT: 18
Note that a large chunk of the improvement fixing regressions caused by
switching to 1-bit Booleans. Previously, our ability to optimize D3D
booleans was improved by using the D3D representation directly in NIR.
Now that NIR does 1-bit bools, we need a few more optimizations.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>