This commit completely rewrites the way we extract a structured CFG from
SPIR-V. The new approach is different in a few ways:
1. It does a breadth-first search instead of depth-first. This means
that we've visited the merge node for a construct before we visit
any of the nodes inside the construct. This makes it easier to
validate things like loop and switch nesting.
2. We record more information in the CFG. Earlier commits added a
parent pointer to vtn_cf_node but we now record all of the merge and
other special blocks for each CFG node. This lets us validate
things more precisely.
3. It makes heavy use of merge blocks for walking the CFG. Previously,
we sort of used them as hints for trying to guess the CFG structure
but things got dicey whenever a merge was missing. We had some
heuristics for how to handle short-circuiting if statements but it
was a bunch of special cases.
Now, we make them a fundamental part of walking the CFG. When we
encounter a control-flow construct, we add the body components of
the construct to the BFS work list and then jump to the merge block
if one exists to continue scanning the current CFG nesting level.
If no merge block exists, we assume that means that control-flow
never re-converges in a normal way and that the only way to get back
to normality is with a direct jump such as a loop break or continue.
This should make things far more robust when trying to deal with the
more creative placement (or lack thereof) of merge instructions.
Reviewed-by: Alan Baker <alanbaker@google.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3820>
Closes: #2760
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4446>
When we originally wrote spirv_to_nir we didn't have a good scalar value
union to handily use so we rolled our own thing for spec constants. Now
that we have nir_const_value, we can use that and simplify a bunch of
the spec constant logic.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4675>
This fixes find_and_update_named_uniform_storage() for subroutines
and also updates num_shader_uniform_components for non opaque
uniforms.
The following patch will ensure this type of bug won't happen again.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4721>
nir_opt_load_store_vectorize has an is_strided_vector() function that
looks for types with weird explicit strides. It does so by comparing
the explicit stride against the type-size-derived typical stride.
This had a subtle bug. Simple vector types (vec2/3/4) have no explicit
stride, so glsl_get_explicit_stride() returns 0. This never matches the
typical stride for a vector, so is_strided_vector() would return true
for basically any vector type, causing the vectorizer to bail.
I found this by looking at a compute shader with scalar SSBO loads at
offsets 0x220, 0x224, 0x228, 0x22c. nir_opt_load_store_vectorize would
properly vectorize the first two into a vec2 load, but would refuse to
extend it to a vec3 and ultimately vec4 load because is_strided_vector()
saw a vec2 and freaked out.
Neither ACO nor ANV do load/store vectorization before lowering derefs,
so this shouldn't affect them. However, I'd like to fix this bug to
avoid the trap for anyone who decides to in the future. In a branch
where anv used this lowering, this cut an additional 38% of the send
messages in the shader by properly vectorizing more things.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4255>
Version 4.4 of the GLSL spec changed the definition of noise*() to
always return zero and earlier versions of the spec allowed zero as a
valid implementation.
All drivers, as far as I can tell, unconditionally call lower_noise()
today which turns ir_unop_noise into zero. We've got a 10-year-old
comment in there saying "In the future, ir_unop_noise may be replaced by
a call to a function that implements noise." Well, it's the future now
and we've not yet gotten around to that. In the mean time, the GLSL
spec has made doing so illegal.
To make things worse, we then pretend to handle the opcode in
glsl_to_nir, ir_to_mesa, and st_glsl_to_tgsi even though it should never
get there given the lowering. The lowering in st_glsl_to_tgsi defines
noise*() to be 0.5 which is an illegal implementation of the noise
functions according to pre-4.4 specs. We also have opcodes for this in
NIR which are never used because, again, we always call lower_noise().
Let's just kill the whole opcode and make builtin_builder.cpp build a
bunch of functions that just return zero.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4624>
If we have a clip dist float[1] compact followed by a tess factor
float[2] we don't want to overlap them, but the partial check
only happens for non-compact vars.
This fixes some issues seen with my sw vulkan layer with
dEQP-VK.clipping.user_defined.clip_distance*
v2: v1 failed with clip/cull mixtures, since in that
case the cull has a location_frac to follow after the clip
so only reset if we get a location_frac of 0 in a subsequent
clip var
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4635>
After the decorations of a variable are evaluated, propagate the
access flag to the associated vtn_pointer. This was done when
creating the pointer but at that point there was no access flags for
the variable.
Inline the pointer creation to make this point clearer, in isolation
the helper made the impression that the value was being propagated.
Issue found by Ken.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4620>
This helps us avoid moving the movs outside if branches when there
src can't be scalarized.
For example it avoids:
vec4 32 ssa_7 = tex ssa_6 (coord), 0 (texture), 0 (sampler),
if ... {
r0 = imov ssa_7.z
r1 = imov ssa_7.y
r2 = imov ssa_7.x
r3 = imov ssa_7.w
...
} else {
...
if ... {
r0 = imov ssa_7.x
r1 = imov ssa_7.w
...
else {
r0 = imov ssa_7.z
r1 = imov ssa_7.y
...
}
r2 = imov ssa_7.x
r3 = imov ssa_7.w
}
...
vec4 32 ssa_36 = vec4 r0, r1, r2, r3
Becoming something like:
vec4 32 ssa_7 = tex ssa_6 (coord), 0 (texture), 0 (sampler),
r0 = imov ssa_7.z
r1 = imov ssa_7.y
r2 = imov ssa_7.x
r3 = imov ssa_7.w
if ... {
...
} else {
if ... {
r0 = imov r2
r1 = imov r3
...
else {
...
}
...
}
While this is has a smaller instruction count it requires more work
for the same result. With more complex examples we can also end up
shuffling the registers around in a way that requires more registers
to use as temps so that we don't overwrite our original values along
the way.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4636>
We can't move them later as we could move them into non-uniform
control flow, but moving them earlier should be fine.
This helps avoid a bunch of spilling in unigine shaders due to
moving the tex instructions sources earlier (outside if branches)
but not the instruction itself.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4636>
Classically, global code motion is also a dead code pass. However, in
the initial implementation, the decision was made to place every
instruction and let conventional DCE clean up the dead ones. Because
any uses of a dead instruction are unreachable, we have no late block
and the dead instructions are always scheduled early. The problem is
that, because we place the dead instruction early, it pushes the
placement of any dependencies of the dead instruction earlier than they
may need to be placed. In order prevent dead instructions from
affecting the placement of live ones, we need to delete them.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4636>
We are about to adjust our instruction block assignment algorithm and we
will want to know the current block that the instruction lives in. In
order to allow for this, we can't overwrite nir_instr::block in the
early scheduling pass.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4636>
Inside nested flow control, nir_lower_returns inserts predicated breaks
in the outer block. However, it would omit doing this if the remainder
of the outer block (after the inner block) was empty. This is not
correct in the case of loops, as execution just wraps back around to the
start of the loop, so this change doesn't skip the predication inside
loops.
Fixes: 79dec93ead (nir: Add return lowering pass)
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2724
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4603>
