Use the same code as the pointer type. It works with all types and works
with any vector, but we need to handle i1 and v3i16 as special cases,
otherwise LLVM fails when it sees them. The previous code only extracted
the first component, which is not what we want.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28607>
I'm not sure why zink would want dri2 here instead of kopper,
I'm sure it's some side effect of something else, let zink
use the kopper paths.
This fixes:
dEQP-GL45-ES3.info.vendor
on zink on nvk with GL cts using EGL.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Fixes: 12a47b84b7 ("egl/dri2: trigger drawable invalidation from surface queries for zink")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28707>
We can use any swizzle to access it, so if the inline constant is in
ADD,try to have the swizzle same as the second channel so that we can
potentially use presubtract later.
RV530:
total instructions in shared programs: 129366 -> 129306 (-0.05%)
instructions in affected programs: 10191 -> 10131 (-0.59%)
helped: 56
HURT: 12
total presub in shared programs: 8421 -> 8791 (4.39%)
presub in affected programs: 1523 -> 1893 (24.29%)
helped: 0
HURT: 102
total temps in shared programs: 17561 -> 17562 (<.01%)
temps in affected programs: 47 -> 48 (2.13%)
helped: 2
HURT: 4
total lits in shared programs: 3021 -> 3179 (5.23%)
lits in affected programs: 427 -> 585 (37.00%)
helped: 0
HURT: 37
total cycles in shared programs: 197845 -> 197512 (-0.17%)
cycles in affected programs: 18949 -> 18616 (-1.76%)
helped: 50
HURT: 19
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28428>
We want to move it after rc_remove_constants because we can't reswizzle
presubtract sources and we need to pack multiple uniforms into vec4
slots to be able to run glamor shaders on R400.
This mostly works OK, the detection logic just needs to recognize the
case when 2.0 was converted to inline constant.
RV530:
total instructions in shared programs: 129384 -> 129366 (-0.01%)
instructions in affected programs: 2078 -> 2060 (-0.87%)
helped: 16
HURT: 7
total presub in shared programs: 8404 -> 8421 (0.20%)
presub in affected programs: 423 -> 440 (4.02%)
helped: 19
HURT: 25
total omod in shared programs: 410 -> 429 (4.63%)
omod in affected programs: 10 -> 29 (190.00%)
helped: 0
HURT: 13
total temps in shared programs: 17560 -> 17561 (<.01%)
temps in affected programs: 34 -> 35 (2.94%)
helped: 3
HURT: 3
total lits in shared programs: 3006 -> 3021 (0.50%)
lits in affected programs: 267 -> 282 (5.62%)
helped: 13
HURT: 20
total cycles in shared programs: 197840 -> 197845 (<.01%)
cycles in affected programs: 4159 -> 4164 (0.12%)
helped: 13
HURT: 13
R420:
total instructions in shared programs: 103810 -> 103762 (-0.05%)
instructions in affected programs: 4176 -> 4128 (-1.15%)
helped: 29
HURT: 1
total presub in shared programs: 2724 -> 2878 (5.65%)
presub in affected programs: 478 -> 632 (32.22%)
helped: 6
HURT: 40
total omod in shared programs: 411 -> 421 (2.43%)
omod in affected programs: 3 -> 13 (333.33%)
helped: 0
HURT: 7
total temps in shared programs: 16841 -> 16852 (0.07%)
temps in affected programs: 814 -> 825 (1.35%)
helped: 5
HURT: 17
total cycles in shared programs: 162397 -> 162361 (-0.02%)
cycles in affected programs: 5746 -> 5710 (-0.63%)
helped: 30
HURT: 3
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28428>
And don't use presubtract in the first place instead of relying on the
dataflow swizzle pass to sort things up later.
Just two shaders in shader-db hits this, but this patch prevents some
regressions in dEQP when we later move presubtract after the dataflow
swizzle pass.
RV420:
presub helped: shaders/chromeos/24.shader_test FS: 6 -> 2 (-66.67%)
presub helped: shaders/chromeos/21.shader_test FS: 6 -> 2 (-66.67%)
cycles helped: shaders/chromeos/24.shader_test FS: 79 -> 78 (-1.27%)
cycles helped: shaders/chromeos/21.shader_test FS: 47 -> 46 (-2.13%)
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28428>
The current helper anv_physical_device_bindless_heap_size()
artificially limited the surface heap size on DG2+ to 128MB. The HW is
actually 4GB capable, but we have workaround requiring to overlap the
dynamic state heap with the bindless surface state heap.
The actual limit comes from our virtual address space setup. It is
different between descriptor buffers and regular descriptors.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: fe037dec6e ("anv: expose VK_EXT_descriptor_buffer")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27873>
Christian Gmeiner