This should get texture gather working on gen8+ and mostly working on gen7.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
While SPIR-V technically doesn't support "old style" shadow, the
shadow-compare gather instruction does return a vec4 so we need to be able
to set the old_style_shadow bit in NIR.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
We can't get an lod with txf_ms and SPIR-V considers textureGrad to be an
explicit-LOD texturing instruction.
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
This allows the load-propagation pass to swap the sources in presence
of immediate values.
Maxwell (GM107):
total instructions in shared programs :1928187 -> 1927634 (-0.03%)
total gprs used in shared programs :330741 -> 330154 (-0.18%)
total local used in shared programs :28032 -> 28032 (0.00%)
local gpr inst bytes
helped 0 271 425 425
hurt 0 0 194 194
Fermi (GF114):
total instructions in shared programs :2334474 -> 2333829 (-0.03%)
total gprs used in shared programs :380934 -> 380215 (-0.19%)
total local used in shared programs :33304 -> 33264 (-0.12%)
local gpr inst bytes
helped 5 314 521 521
hurt 0 4 195 195
No regressions on GM107 and GF114 with full piglit.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
There are 2 uses:
- Asynchronous flushing for multithreaded drivers.
- Return a fence without flushing (mid-command-buffer fence). The driver
can defer flushing until fence_finish is called.
This is required to make Bioshock Infinite faster, which creates
1000 fences (flushes) per frame.
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Rob Clark <robdclark@gmail.com>
subroutine variables are to be used just in the way functions are
called. Although the spec doesn't say it explicitely, this means that
these variables are not to be used in any other way than those left
for function calls. Therefore, a comparison between 2 subroutine
variables should also cause a compilation error.
From The OpenGL® Shading Language 4.40, page 117:
" To use subroutines, a subroutine type is declared, one or more
functions are associated with that subroutine type, and a
subroutine variable of that type is declared. The function
currently assigned to the variable function is then called by
using function calling syntax replacing a function name with the
name of the subroutine variable. Subroutine variables are
uniforms, and are assigned to specific functions only through
commands (UniformSubroutinesuiv) in the OpenGL API."
From The OpenGL® Shading Language 4.40, page 118:
" Subroutine uniform variables are called the same way functions
are called. When a subroutine variable (or an element of a
subroutine variable array) is associated with a particular
function, all function calls through that variable will call that
particular function."
Fixes GL44-CTS.shader_subroutine.subroutines_cannot_be_assigned_float_int_values_or_be_compared
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Since 7f53fead5c we treat every location as using all
four components so we only need special handling for
doubles when they cross multiple locations.
This fixes a crash in GL45-CTS.enhanced_layouts.varying_locations
where the outputs array would overflow when a dmat2 was stored at
the max varying location i.e 30.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This fixes a regression introduced in
1da704a94c because the offset has moved
from 0x180 to 0x1a0, and the macros have to be re-compiled.
Fixes: 1da704a ("nvc0: increase the tex handles area size in the driver")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This fixes a regression introduced in
1da704a94c because the offset has moved
from 0x600 to 0x620, and the kernels used for reading MP perf counters
have to be re-assembled.
This also fixes amd_performance_monitor_measure piglit.
Fixes: 1da704a ("nvc0: increase the tex handles area size in the driver")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
The GL_EXT_texture_format_BGRA8888 extension specification defines a
GL_BGRA_EXT unsized internal format (which is a little odd - usually
BGRA is a pixel transfer format). The extension is written against
the ES 1.0 specification, so it's a little hard to map, but I believe
it's effectively adding it to the table used here, so we should allow
it here as well.
Note that GL_EXT_texture_format_BGRA8888 is always enabled (dummy_true),
so we don't need to check if it's enabled here.
This fixes mipmap generation in Skia and ChromeOS.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
References: https://bugs.chromium.org/p/chromium/issues/detail?id=630371
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reported-by: Stéphane Marchesin <marcheu@chromium.org>
Cc: mesa-stable@lists.freedesktop.org
Commit 52e75dcb8c made nir_lower_io
start using nir_intrinsic_set_base instead of writing const_index[0]
directly. However, those intrinsics apparently don't /have/ a base,
so this caused assert failures.
However, the old code was happily setting non-existent const_index
fields, so it was pretty bogus too.
Jason pointed out that load_shared and store_shared have a base,
and that the i965 driver uses that field. So presumably atomics
should have one as well, so that loads/stores/atomics all refer
to variables with consistent addressing.
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
We can still do packing we just need to get the packing type from the consumer
rather than the producer.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97033
We always resort to the pull model for instanced GS inputs. So, we'd
better include the VUE handles, or else we can't actually pull anything.
Ian reports that on his branch with OES_geometry_shader enabled,
this fixes a bunch of dEQP-GLES31.functional.geometry_shading tests::
- instanced.draw_2_instances_geometry_2_invocations
- instanced.draw_2_instances_geometry_8_invocations
- instanced.draw_4_instances_geometry_2_invocations
- instanced.draw_4_instances_geometry_8_invocations
- instanced.draw_8_instances_geometry_2_invocations
- instanced.draw_8_instances_geometry_8_invocations
- instanced.geometry_2_invocations
- instanced.geometry_32_invocations
- instanced.geometry_8_invocations
- instanced.geometry_max_invocations
- instanced.geometry_output_different_2_invocations
- instanced.geometry_output_different_32_invocations
- instanced.geometry_output_different_8_invocations
- instanced.geometry_output_different_max_invocations
- instanced.invocation_output_vary_by_attribute
- instanced.invocation_output_vary_by_texture
- instanced.invocation_output_vary_by_uniform
- query.primitives_generated_instanced
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
Here we create a new output_generic_reg array with the ability to
store the dst_reg for each component of user defined varyings.
This is needed as the previous code only stored the dst_reg based
on the varying location which meant packed varyings would overwrite
each other.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
For example where n=3 first_component=1 this will give us
0xE (WRITEMASK_YZW).
V2:
Add assert to check first component is <= 4 (Suggested by Ken)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
This will be used to swizzle components to the beginning or end
of the vector based on the component layout qualifier and whether
we are doing a load or store.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
This makes sure we give the correct driver location
for doubles when using component packing. Specifically
it handles packing a dvec3 with a double which is the
only packing scenario allowed which spans across two
locations.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Here we use the component qualifier (which is the first component)
as an offset when loading output varyings.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Rather than trying to work out the total number of components
used at a location we simply treat all outputs as vec4s. This
removes the need for complex code looping over varyings to match
packed locations and the need for storing the total number of
components used at each location.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
We will use this for output varyings. To make component
packing simpler we will just treat all varyings as vec4s.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The build was failing because the official CL headers have a few defines,
like: # define cl_khr_gl_sharing 1
Which have the same name as some class members of clang's OpenCLOptions class.
If we include the cl headers first, this breaks the build because the member
names of this class are replaced by the literal 1.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Vedran Miletić <vedran@miletic.net>
clang commit r275822 removed unnecessary includes from header files,
so we now need to explicitly include clang/Lex/PreprocessorOptions.h
v2:
- Use <> instead of "" for the include path.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Vedran Miletić <vedran@miletic.net>
TCS/TES/GS and now FS all handle these in stage-specific functions.
CS don't have inputs, so VS was the only one left using this code.
Move it to the VS-specific function for clarity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We no longer use this message. As far as I can tell, it's fairly
useless - the equivalent information is provided in the payload.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This eliminates the need to walk the list of input variables, recurse
into their types (via logic largely redundant with nir_lower_io), and
interpolate all possible inputs up front. The backend no longer has
to care about variables at all, which eliminates complications from
trying to pack multiple variables into the same location. Instead,
each intrinsic specifies exactly what's needed.
This should unblock Timothy's work on GL_ARB_enhanced_layouts.
Each load_interpolated_input intrinsic corresponds to PLN instructions,
while load_barycentric_at_* intrinsics correspond to pixel interpolator
messages. The pixel/centroid/sample barycentric intrinsics simply refer
to payload fields (delta_xy[]), and don't actually generate any code.
Because we use a single intrinsic for both centroid-qualified variables
and interpolateAtCentroid(), they become indistinguishable. We stop
sending pixel interpolator messages for those, and instead use the
payload provided data, which should be considerably faster.
On Broadwell:
total instructions in shared programs: 9067751 -> 9067570 (-0.00%)
instructions in affected programs: 145902 -> 145721 (-0.12%)
helped: 422
HURT: 209
total spills in shared programs: 2849 -> 2899 (1.76%)
spills in affected programs: 760 -> 810 (6.58%)
helped: 0
HURT: 10
total fills in shared programs: 3910 -> 3950 (1.02%)
fills in affected programs: 617 -> 657 (6.48%)
helped: 0
HURT: 10
LOST: 3
GAINED: 3
The differences mostly appear to be slight changes in MOVs.
v2: Use nir_shader_compiler_options::use_interpolated_input_intrinsics
flag rather than passing it directly to nir_lower_io. Use the
unreachable() macro rather than assert in one place. (Review
feedback from Chris Forbes.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Currently, i965 interpolates all FS inputs at the top of the program.
This has advantages and disadvantages, but I'd like to keep that policy
while reworking this code. We can consider changing it independently.
The next patch will make the compiler generate PLN instructions "on the
fly", when it encounters an input load intrinsic, rather than doing it
for all inputs at the start of the program.
To emulate this behavior, we introduce an ugly pass to move all NIR
load_interpolated_input and payload-based (not interpolator message)
load_barycentric_* intrinsics to the shader's start block.
This helps avoid regressions in shader-db for cases such as:
if (...) {
...load some input...
} else {
...load that same input...
}
which CSE can't handle, because there's no dominance relationship
between the two loads. Because the start block dominates all others,
we can CSE all inputs and emit PLNs exactly once, as we did before.
Ideally, global value numbering would eliminate these redundant loads,
while not forcing them all the way to the start block. When that lands,
we should consider dropping this hacky pass.
Again, this pass currently does nothing, as i965 doesn't generate these
intrinsics yet. But it will shortly, and I figured I'd separate this
code as it's relatively self-contained.
v2: Dramatically simplify pass - instead of creating new instructions,
just remove/re-insert their list nodes (suggested by Jason Ekstrand).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
