If we have NIR such as:
32x4 %48 = @load_vulkan_descriptor (%47) (desc_type=SSBO)
32x4 %76 = deref_cast (tint_symbol_11 *)%48 (ssbo tint_symbol_11) (ptr_stride=0, align_mul=4, align_offset=0)
32x4 %77 = deref_struct &%76->tint_symbol_10 (ssbo int) // &((tint_symbol_11 *)%48)->tint_symbol_10
A single nir_rematerialize_deref_in_use_blocks() will rematerialize the
deref_struct and then it's deref_cast. However,
nir_foreach_instr_reverse_safe is not safe if the next iteration's
instruction is removed. This can result in the instruction loop exiting
and the load_vulkan_descriptor never having an LCSSA phi.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Fixes: 439e8c42cc ("nir/lcssa: Fix rematerializing derefs")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11770
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32225>
iabs(a) is not positive if "a" is the minimum signed value, so this is
incorrect in that case for some values of "b".
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Fixes: 2b76de9b5d ("nir/algebraic: Add a couple optimizations for iabs and ishr")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32157>
The code wants the number of components used by the variable in the
current attribute slot, not the total number of components.
For e.g. a 4x3 matrix, glsl_get_components() returns 12, leading to the
following error reported by AddressSanitizer:
```
Test case 'dEQP-VK.tessellation.shader_input_output.cross_invocation_per_patch_mat4x3'..
../src/compiler/nir/nir_lower_io_to_vector.c:265:16: runtime error: index 4 out of bounds for type 'nir_variable *[4]'
```
Fixes: 5ef2b8f1f2 ("nir: Add a pass for lowering IO back to vector when possible")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32193>
If the hardware does not support INSTANCE_INDEX natively, it will be
lowered to load_instance_id + base_instance. Otherwise, INSTANCE_ID
will be lowered to load_instance_index - base_instance.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32158>
The semantics of this newly introduced system value match
Vulkan's InstanceIndex exactly, and are equivalent to
instance_id + base_instance.
Some hardware, such as Mali Valhall or later, only provides
instance id offset by base_instance. Introducing a new system
value to represent this, rather than handling the mismatch
when lowering to BIR lets us use NIR to eliminate redundant
arithmetic that would follow from mismatched semantics, e.g.
instance_id could be lowered to instance_index - base_instance,
so expressions such as instance_id + base_instance would be
optimized to a simple instance_index.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32158>
A recent algebraic opt made a function that used to inline
with llvmpipe CL not inline anymore. However that function
has a barrier in it.
Handling barriers from inside a callstack is hard for llvmpipe
coroutines, so just force functions with barriers to be inlined.
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32204>
We have to honor drivers when they say that different interpolation
qualifiers can't be mixed in the same vec4, indicated by
nir_io_has_flexible_input_interpolation_except_flat not being set.
This is a prerequisite for enabling nir_opt_varyings for all drivers.
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32174>
This was enabled by default in nir_opt_varyings, but vc4 can't handle
when shader outputs write Y but not X. Add an option for it and enable
it only for the driver that benefits from it.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32174>
The main difference between load_global and load_global_constant is that
the latter can be reordered arbitrarily. If the access being lowered is
already tagged as being reorderable, then we can preserve that by using
the load_global_constant intrinsics instead of load_global. This gives
us more flexibility.
On Intel, this lets us use the load_global_constant_uniform_block_intel
intrinsic for doing convergent block loads in more cases. This nets us
significant reductions in spill/fills: Borderlands 3 on Lunarlake sees
spills/fills reduced by 53%. Alchemist sees a 13% reduction.
Improves performance of Borderlands 3 DX12 on Intel Battlemage by
around 44%. Improves Hogwarts Legacy by around 14%.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31995>
When (off_const > max) there is a wrap around uint when calling
try_extract_const_addition.
Exit early since folding doesn't make sense in this case.
Cc: mesa-stable
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32118>
This instruction can be used as a breakpoint in shaders to enter a
trap if supported by the driver. It will be used to handle
NonSemantic.DebugBreak in SPIR-V.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32061>
ishl can wrap, amul cannot. so we need amul in the backend, or otherwise we
would need to introduce an ashl opcode instead. that doesn't seem better.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31964>
Some drivers generally need int64 lowered, but prefer to do this lowering
themselves late, to have a chance to optimize targeted int64 patterns before
lowering the rest. This isn't currently possible since nir_lower_int64 takes no
options except what's const* in the shader, and frontends call nir_lower_int64
before passing the shader off to the driver. Add an option to defer int64
lowering. This is a bit ugly but the alternative is replumbing nir_lower_int64's
option handling cross-tree and no-thank-you-not-right-now.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31964>
This only worked when "a" was 16-bit because a pattern above replaced the
shift.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31762>
It only supported scalar intrinsics because it was written before
nir_opt_vectorize_io existed. The introduction of nir_opt_vectorize_io
exposes this issue. The direct path has been tested. The indirect path
hasn't. That's fine because if we see a CLIP_DIST failure with indirect
in the future, this pass is likely the cause.
This is a prerequisite for enabling nir_opt_varyings for all gallium
drivers.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31994>
