| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
| |
Previously, we would create temporary variables and fill them out.
Instead, we create as many function parameters as we need and pass them
through as SSA defs.
Reviewed-by: Iago Toral Quiroga <[email protected]>
|
|
|
|
|
|
|
| |
Instead of initializing them manually, just use the type that we already
have sitting there.
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, we would always pull the bit size from the destination which
is wrong for opcodes like nir_ilt where the sources are variable-sized
but the destination is a fixed size. We were getting lucky before
because nir_op_ilt returns a 32-bit value and basically everyone who
uses spec constants uses 32-bit ones.
Cc: [email protected]
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
| |
Reviewed-by: Ian Romanick <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
This extension adds two new decorations which carry meaning only for
HLSL shaders. They are expected to be handled by higher level layers
and can be ignored by implementations. However, it does save the client
a bit of work if the implementation safely ignores them instead of the
client having to strip them out of the SPIR-V in order for it to be
valid.
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
|
|
|
|
| |
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
|
|
|
|
|
|
| |
This corresponds to commit 801cca8104245c07e8cc532 on GitHub.
Acked-by: Bas Nieuwenhuizen <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
v2: - change how the access qualifiers are accumulated
v3: - duplicate members in struct_member_decoration_cb()
- handle access qualifiers on variables
- remove access qualifiers handling in _vtn_variable_load_store()
- fix setting access qualifiers on type->array_element
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]
|
|
|
|
|
|
|
|
| |
Otherwise, they are removed during NIR linking or in some
lowering passes.
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
| |
It makes way more sense for it to live there with the rest of function
handling.
Reviewed-by: Iago Toral Quiroga <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This commit expands the current memory access enum to contain the extra
two bits provided for images. We choose to follow the SPIR-V convention
of NonReadable and NonWriteable because readonly implies that you *can*
read so readonly + writeonly doesn't make as much sense as NonReadable +
NonWriteable.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Instead of requiring 4 components, this allows them to potentially use
fewer. Both the SPIR-V and GLSL paths still generate vec4 intrinsics so
drivers which assume 4 components should be safe. However, we want to
be able to shrink them for i965.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GLSL has gl_VertexID which is supposed to be non-zero-based.
SPIR-V has both VertexIndex and VertexId builtins whose meanings are
defined by the APIs.
Vulkan defines VertexIndex as being non-zero-based. In Vulkan VertexId
and InstanceId have no meaning and are pretty much just reserved for
OpenGL at this point.
GL_ARB_spirv removes VertexIndex and defines VertexId to be the same
as gl_VertexId (which is also non-zero-based).
Previously in Mesa it was treating VertexIndex as non-zero-based and
VertexId as zero-based, so it was breaking for GL. This behaviour was
apparently based on Khronos bug 14255. However that bug doesn’t seem
to have made a final decision for VertexId.
Assuming there really is no other definition for VertexId for Vulkan
it seems better to just make them both have the same value.
v2: update comment and commit descriptions, based on Jason Ekstrand
explanation of the meaning/rationale behind all those builtins
(Jason)
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
info.gs.output_primitive was already being filled. Not sure why this
is not needed on Vulkan, but we found to be needed for
ARB_gl_spirv. Specifically, this is needed to get the following test
passing:
KHR-GL45.gl_spirv.spirv_validation_builtin_variable_decorations_test
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now that all the build scripts are compatible with both Python 2 and 3,
we can flip the switch and tell Meson to use the latter.
Since Meson already depends on Python 3 anyway, this means we don't need
two different Python stacks to build Mesa.
Signed-off-by: Mathieu Bridon <[email protected]>
Reviewed-by: Eric Engestrom <[email protected]>
Reviewed-by: Dylan Baker <[email protected]>
|
|
|
|
| |
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
| |
It looks like it was previously taking the SPIR-V instruction number
directly instead of looking up the constant value.
v2: use vtn_constant_value helper (Jason)
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
From SPIR-V 1.0 spec, section 3.20, "Decoration":
"Stream
Apply to an object or a member of a structure type. Indicates the
stream number to put an output on."
Note the "or", so that means that it is allowed for both a full struct
or a membef or a struct (although the wording is not really ideal, and
somewhat error-prone, imho).
We found this with some Geometry Streams tests for ARB_gl_spirv, where
the full gl_PerVertex is assigned Stream 0 (default value on OpenGL
for gl_PerVertex).
So this commit allows structs to have this Decoration, and sets the
stream at the nir variable if needed.
Signed-off-by: Neil Roberts <[email protected]>
Signed-off-by: Alejandro Piñeiro <[email protected]>
v2: squash two Decoration Stream patches (Jason)
Reviewed-by: Jason Ekstrand <[email protected]>
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
| |
These set the new explicit XFB members on nir_variable.
This is needed to support ARB_gl_spirv, as Vulkan doesn't support
transform feedback.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
| |
This just sets has_transform_feedback_varyings on the shader.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
also move some of the GLSL builtins over we will need for implementing
some OpenCL builtins
v2: replace NIR_IMM_FP by nir_imm_floatN_t in ported code
fix up changes caused by swizzle rework
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Lightly edited to be valid 'C' code.
Is there a bug open to fix this upstream?
Acked-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Allow the capability to be exposed, and convert the new execution mode
into fs state.
Signed-off-by: Ilia Mirkin <[email protected]>
Reviewed-by: Lionel Landwerlin <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes warning:
../../src/compiler/spirv/vtn_variables.c: In function ‘var_decoration_cb’:
../../src/compiler/spirv/vtn_variables.c:1400:12: warning: ‘is_vertex_input’ may be used uninitialized in this function [-Wmaybe-uninitialized]
bool is_vertex_input;
^~~~~~~~~~~~~~~
The code used to set is_vertex_input in all possible codepaths, but
after 23edc5b1ef3 "spirv: translate default-block uniforms" the
compiler isn't sure all codepaths will initialize the variable.
Reviewed-by: Anuj Phogat <[email protected]>
|
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
OpenCL knows vector of size 8 and 16.
v2: rebased on master (nir_swizzle rework)
rework more declarations with nir_component_mask_t
adjust print_var_decl
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
For one thing, the NIR opcodes for image load/store always take and
return a vec4 value regardless of the image type. We need to fix up
both the source and destination to handle it. For another thing, we
weren't actually setting up a destination in the OpAtomicLoad case.
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Cc: [email protected]
|
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
There are no fixed sized array arguments in C, those are simply pointers
to unsized arrays and as the size is passed in anyway, just rely on that.
where possible calls are replaced by nir_channel and nir_channels.
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
| |
Reviewed-by: Karol Herbst <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
| |
Otherwise nir_validate may complain about 8 bit floats, which do not exist.
Reviewed-by: Karol Herbst <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
Signed-off-by: Karol Herbst <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If the glsl is something like this:
in vec4 some_input;
interpolateAtCentroid(some_input[idx])
then it now gets generated as if it were:
interpolateAtCentroid(some_input)[idx]
This is necessary because the index will get generated as a series of
nir_bcsel instructions so it would no longer be an input variable. It
is similar to what is done for GLSL in ca63a5ed3e9efb2bd645b42.
Although I can’t find anything explicit in the Vulkan specs to say
this should be allowed, the SPIR-V spec just says “the operand
interpolant must be a pointer to the Input Storage Class”, which I
guess doesn’t rule out any type of pointer to an input.
This was found using the spec/glsl-4.40/execution/fs-interpolateAt*
Piglit tests with the ARB_gl_spirv branch.
Signed-off-by: Neil Roberts <[email protected]>
Signed-off-by: Alejandro Piñeiro <[email protected]>
v2: update after nir_deref_instr land on master. Implemented by
Alejandro Piñeiro. Special thanks to Jason Ekstrand for guidance
at the new nir_deref_instr world.
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
| |
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
| |
Updates headers and grammar to ff684ffc6a35d2a58f0f63108877d0064ea33feb
Acked-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
From the SPIR-V 1.0 specification, section 3.32.18, "Atomic
Instructions":
"OpAtomicIDecrement:
<skip>
The instruction's result is the Original Value."
However, we were implementing it, for uniform atomic counters, as a
pre-decrement operation, as was the one available from GLSL.
Renamed the former nir intrinsic 'atomic_counter_dec*' to
'atomic_counter_pre_dec*' for clarification purposes, as it implements
a pre-decrement operation as specified for GLSL. From GLSL 4.50 spec,
section 8.10, "Atomic Counter Functions":
"uint atomicCounterDecrement (atomic_uint c)
Atomically
1. decrements the counter for c, and
2. returns the value resulting from the decrement operation.
These two steps are done atomically with respect to the atomic
counter functions in this table."
Added a new nir intrinsic 'atomic_counter_post_dec*' which implements
a post-decrement operation as required by SPIR-V.
v2: (Timothy Arceri)
* Add extra spec quotes on commit message
* Use "post" instead "pos" to avoid confusion with "position"
Signed-off-by: Antia Puentes <[email protected]>
Signed-off-by: Alejandro Piñeiro <[email protected]>
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Capability that informs if atomic counters are supported. From SPIR-V
1.0 spec, section 3.7, "Storage Class", item 10 from table:
(Column "Storage Class"):
"AtomicCounter For holding atomic counters. Visible across all
functions of the current invocation. Atomic counter-specific
memory."
(Column "Required Capability"):
"AtomicStorage"
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
| |
So renamed to a more general vtn_handle_atomics
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
| |
This is convenient when dealing with atomic counter uniforms. The
alternative would be doing that at vtn_handle_atomics.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When constructing NIR if we have a SPIR-V uint variable and the
storage class is SpvStorageClassAtomicCounter, we store as NIR's
glsl_type an atomic_uint to reflect the fact that the variable is an
atomic counter.
However, we were tweaking the type only for atomic_uint scalars, we
have to do it as well for atomic_uint arrays and atomic_uint arrays of
arrays of any depth.
Signed-off-by: Antia Puentes <[email protected]>
Signed-off-by: Alejandro Piñeiro <[email protected]>
v2: update after deref patches got pushed (Alejandro Piñeiro)
v3: simplify repair_atomic_type (suggested by Timothy Arceri, included
on the patch by Alejandro)
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GLSL types differentiates uint from atomic uint. On SPIR-V the type is
uint, and the variable has a specific storage class. So we need to
tweak the type based on the storage class.
Ideally we would like to get the proper type at vtn_handle_type, but
we don't have the storage class at that moment.
We tweak only the nir type, as is the one that really requires it.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Also initialize it on var_decoration_cb
This is equivalent to nir_variable.offset, used to store the location
an atomic counter is stored at.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
| |
Atomic Counters are uniforms per spec.
Reviewed-by: Timothy Arceri <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit completely reworks function calls in NIR. Instead of having
a set of variables for the parameters and return value, nir_call_instr
now has simply has a number of sources which get mapped to load_param
intrinsics inside the functions. It's up to the client API to build an
ABI on top of that. In SPIR-V, out parameters are handled by passing
the result of a deref through as an SSA value and storing to it.
This virtue of this approach can be seen by how much it allows us to
delete from core NIR. In particular, nir_inline_functions gets halved
and goes from a fairly difficult pass to understand in detail to almost
trivial. It also simplifies spirv_to_nir somewhat because NIR functions
never were a good fit for SPIR-V.
Unfortunately, there is no good way to do this without a mega-commit.
Core NIR and SPIR-V have to be changed at the same time. This also
requires changes to anv and radv because nir_inline_functions couldn't
handle deref instructions before this change and can't work without them
after this change.
Acked-by: Rob Clark <[email protected]>
Acked-by: Bas Nieuwenhuizen <[email protected]>
Acked-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We were only initializing vtn_builder::func for the pre-walk where we
build the CFG. We were only initializing the nir_builder for the later
walk through the instructions even though were were setting b->cursor
for the pre-walk. Let's set both both places so that everything is
consistent. This useful because we handle OpFunctionParameter in the
pre-walk and we're going to need to be able to emit instructions.
Acked-by: Rob Clark <[email protected]>
Acked-by: Bas Nieuwenhuizen <[email protected]>
Acked-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
| |
Acked-by: Rob Clark <[email protected]>
Acked-by: Bas Nieuwenhuizen <[email protected]>
Acked-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
Now that pointers can be derefs and derefs just produce SSA values, we
can convert any pointer to/from SSA.
Acked-by: Rob Clark <[email protected]>
Acked-by: Bas Nieuwenhuizen <[email protected]>
Acked-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, pointers fell into two categories: index/offset for UBOs,
SSBOs, etc. and var + access chain for logical pointers. This commit
adds another logical pointer mode that's deref + access chain.
It's tempting to think that we can just replace variable-based pointers
with deref-based or at least replace the access chain with a deref
chain. Unfortunately, there are a few sticky bits that prevent this:
1) We can't return deref-based pointers from OpVariable because those
opcodes may come outside of a function so there's no place to emit
the deref instructions.
2) We can't always use variable-based pointers because we may not
always know the variable. (We do now, but he upcoming function
rework will take that option away.)
3) We also can't replace the access chain struct with a deref. Due to
the re-ordering we do in order to handle loop continues, the derefs
we would emit as part of OpAccessChain may not dominate their uses.
We normally fix this up with nir_repair_ssa but that generates phi
nodes which we don't want in the middle of our deref chains.
All in all, we have no real better option than to support partial access
chains while also re-emitting the deref instructions on the spot.
Acked-by: Rob Clark <[email protected]>
Acked-by: Bas Nieuwenhuizen <[email protected]>
Acked-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|