The implementation is inspired by
lower_instructions_visitor::dfrexp_sig_to_arith.
This has been tested against the arb_gpu_shader_fp64/fs-frexp-dvec4
test using the ARB_gl_spirv branch.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
So now, during spirv_to_nir, it uses the capability instead of the
extension. Note that we are really doing here is treating
SPV_AMD_gcn_shader as other supported extensions. SPV_AMD_gcn_shader
is not the first SPV extension supported. For example, the capability
draw_parameters infers if the extension SPV_KHR_shader_draw_parameters
is supported or not.
This could be seen as counter-intuitive, and that it would be easier
to define which extensions are supported, and based our checks on
that, but we need to take into account that some capabilities are
optional from core, and others came from new extensions.
Also this commit would make the implementation of ARB_spirv_extensions
easier.
v2: AMD_gcn_shader capability renamed to gcn_shader (Daniel Schürmann)
Reviewed-by: Daniel Schürmann <daniel.schuermann@campus.tu-berlin.de>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
We don't need anymore the source and destination's data type, just
their bitsize.
v2:
- Use glsl_get_bit_size () instead (Jason).
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
There are some SPIRV opcodes (like UConvert and SConvert) have some
expectations of the output that doesn't depend on the operands
data type. Generalize the solution of all of them.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
OpenCL kernels also have int8/uint8.
v2: remove changes in nir_search as Jason posted a patch for that
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
The code to handle mat multiplication by a scalar tries to pick either
imul or fmul depending on whether the matrix is float or integer.
However it was doing this by checking whether the base type is float.
This was making it choose the int path for doubles (and presumably
float16s).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Co-authored-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Daniel Schürmann <daniel.schuermann@campus.tu-berlin.de>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Our previous handling of barriers always used the big hammer and didn't
correctly emit memory barriers when specified along with a control
barrier. This commit completely reworks the way we emit barriers to
make things both more precise and more correct.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This capability allows gl_ViewportIndex and gl_Layer to also be used
as outputs in Vertex and Tesselation shaders.
v2: Make conditional to the capability, add gl_Layer, add tesselation
shaders. (Iago)
v3: Don't export to tesselation control shader.
v4: Add Reviewd-by tag.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
The introduction of 16-bit types with VK_KHR_16bit_storages implies that
push constant offsets could be multiple of 2-bytes. Some assertions are
updated so offsets should be just multiple of size of the base type but
in some cases we can not assume it as doubles aren't aligned to 8 bytes
in some cases.
For 16-bit types, the push constant offset takes into account the
internal offset in the 32-bit uniform bucket adding 2-bytes when we access
not 32-bit aligned elements. In all 32-bit aligned cases it just becomes 0.
v2: Assert offsets to be aligned to the dest type size. (Jason Ekstrand)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Range in 16-bit push constants load was being calculated
wrongly using 4-bytes per element instead of 2-bytes as it
should be.
v2: Use glsl_get_bit_size instead of if statement
(Jason Ekstrand)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The SPIR-V parser splits in/out struct variables and creates
a separate variable for each first-level member of the struct.
When the struct variable has an initializer this means that we also
need to split the initializer.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
New generated files from:
bb1e6ff161 ("spirv: Add a prepass to set types on vtn_values")
65fc16c974 ("autotools: set XA versions in configure.ac and configure header file")
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Technically, the GLSLang bug related to this can also affect SSBO writes
where the bool -> uint conversion is missing. However, the only known
shipping application with an old enough version of GLSLang to cause
issues with this is the new DOOM game so we keep the workaround as small
as possible.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104424
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Previously, we were storing a pointer to the vtn_value because we use it
to look up decorations when we create input/output variables. This
works, but it also may be useful to have the id itself so we may as well
store that instead.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that higher levels are enforcing decoration sanity, we don't need
the vtn_asserts here. This function *should* be safe but we still want
a few well-placed regular asserts in case something goes awry.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This reworks the error checking on our generic handling of decorations.
The objective is to validate all of the SPIR-V assumptions we make
up-front and convert redundant checks to compiled-out asserts. The most
important part of this is to ensure that member decorations only occur
on OpTypeStruct and that the member is never out-of-bounds. This way
later code can assume that the member is sane and not have to worry
about OOB array access due to a misplaced OpMemberDecorate.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This is a bit simpler since we have fewer enum values in the case. It's
also a bit more efficient because we're making fewer glsl_get_* calls.
While we're at it, add better type validation.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that vtn_base_type is a real and full base type, we can switch on
that instead of the GLSL base type which is a lot fewer cases in our
switch.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This reverts commit 9702fac68e, which
hangs vulkancts and crucible on all platforms.
The patch is being reverted because it disables continuous integration
testing. The patch from bug 104359 does not apply to master.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104359
When walking over all the cases in a OpSwitch, take in account the bitsize
of the literals to avoid getting wrong cases.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The Talos Principle contains shaders with an OpSelect between two
vectors where the condition is a scalar boolean. This is technically
against the spec bout nir_builder gracefully handles it by splatting
out the condition to all the channels. So long as the condition is a
boolean, just emit a warning instead of failing.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104246
