Commit 24342e499b changed how primitive
topology is handled on Gfx8+ but missed updating the Gfx7.x code.
As a result, tests which previously used topologies like PATCHLIST_3
instead started using bogus ones like LINESTRIP_ADJ. This caused a
GPU hangs in a bunch of Vulkan conformance tests involving tessellation.
This fixes those hangs.
Fixes: 24342e499b ("anv: fix dynamic primitive topology for tess")
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11434>
This has two steps. First, for each range we look at the memory object
and see if it actually needs flushing before we start throwing CLFLUSH
instructions. Second, we look at the whole list of types on device
initialization and decide whether or not we need CLFLUSH at all. The
first part should speed up atom chips a bit since we're currently
CLFLUSHing everything even when we don't need to. The second isn't
needed on most of today's parts because we base it on !has_llc but it is
needed for discrete parts. It's also over-all cleaner.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11364>
HDC Pipeline Flush is the correct method for flushing HDC
pipeline on Gfx12+ HW. Continue using DC Flush for earlier HW.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9834>
Tile Cache flush flushes all Color/Depth values from L3 cache
to memory in Unified Cache mode. This is only required when
CPU access is required.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9834>
On Gfx12+, flushing tile cache ensures color/depth values are
globally visible, but that's expensive. Most operations only
need values to be GT-visible which can be achieved with depth
or rt flush. Remove a bunch of unnecessary Tile Cache flushes.
Fast clears and slow depth clears still require Tile Cache flush.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9834>
This is a gallium driver for the Intel gfx 4-7 GPUs.
It was initially cloned from the iris driver by Ilia Mirkin,
then I ported over large reams of code from i965 until it worked.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11146>
965 and the mesa st disagree on how vertex elements are ordered when
edgeflags are involved. 965 wants them in gl_vert_attrib order,
but gallium supplies the edgeflag as the last vertex element regardless.
This adds a flag which is enabled for gen4/5 to denote that the
edgeflag is at the end. When we reap 965 later we can resolve this
better.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11146>
Now that there's a common NIR pass, there's no point in us doing this in
the back-end anymore. In order to use this pass in i965, we do have to
make one tiny change. Gallium runs the pass after assigning input and
output locations and so needs the pass to respect those locations and
num_inputs. i965, however, runs it before any location assignment or
I/O lowering so we don't care. We do, however, need the pass to succeed
with num_inputs == 0 because we set that later.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11313>
With Yf and Ys tiling, everything is actually four dimensional because
we can have multiple depth or multisampled array slices in the same
tile. This commit just enhances the calculations so they can handle it.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11330>
We need to do this in order to handle Yf and Ys tiling because they use
a four-dimensional tile instead of laying everything out in two
dimensions.
v2 (Jason Ekstrand):
- Update functions added since v1:
- isl_surf_get_image_range_B_tile
- blorp_can_hiz_clear_depth
- get_image_offset_el
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com> (v1)
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11330>
This should drop the CPU overhead of processing buffers on SKL+ by
dropping some of the logic contained in anv_reloc_list_add() whenever we
have enough compile-time information to know we have softpin.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11236>
The relocation list currently serves two purposes. One is for
relocations on older non-softpin platforms. The second is to keep track
of driver-managed BOs which are used by the given command buffer. We
going to need a mechanism to add BOs to the command buffer without doing
a relocation into the batch.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11236>
Whenever we have the GFX_VERx10 macro available, we can make use_softpin
a compile-time thing for everything but Broadwell and Cherryview. This
should save us some CPU cycles especially on SKL+.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11236>
Softpin was added to i915 in
commit 506a8e87d8d2746b9e9d2433503fe237c54e4750
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Tue Dec 8 11:55:07 2015 +0000
drm/i915: Add soft-pinning API for execbuffer
which was included in Linux 4.5. It's been over 5 years so it's
probably reasonable to make it a hard requirement.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Acked-by: Emma Anholt <emma@anholt.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11236>
Mesh and Task shaders can use workgroup memory, so generalize its
handling in anv by moving it from anv_pipeline_compile_cs() to
anv_pipeline_lower_nir().
Update Pipeline Statistics accordingly.
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11230>
Move it out the "cs" sub-struct, since the bit will be used for other
shader stages in the future.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11225>
This drops
-0000000000011e90 R isl_format_layouts
+0000000000008f48 R isl_format_layouts
I think that's about 36k.
Thanks to Jason for suggesting PACKED
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11232>
Be consistent with other usages in Vulkan and SPIR-V, and the recently
added workgroup_size field.
Acked-by: Emma Anholt <emma@anholt.net>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11190>
Since 40e1d798c6, we are now using physical register numbers for
everything which makes it all simpler. In particular, we no longer need
the special case for setting up the payload for SIMD16 on Gen4-5. This
fixes a pile of piglit tests on ILK and similar.
Fixes: 40e1d798c6 "intel/fs: Use ra_alloc_contig_reg_class()..."
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11221>
The format enum space isn't necessarily contiguous so we can't assume
that if it's in the table it's valid. We need to check something.
Fixes: ed6e586562 "intel: properly constify isl_format_layouts"
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11191>
By using the new class type, we don't need to make 1928 different
registers to represent each contigous reg size starting from the actual
128 HW register, or have a mapping between RA regs and HW base regs. With
the number of regs reduced, and the fast q computation when using the new
classes, we no longer need to compute our own q.
This drops the FS RA initialization time on my CFL system from about 1ms to
50us.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9437>
Putting a const char * in the struct means it's a pointer that has to be
resolved at rtld time, which means it can be in .data.rel.ro but not
.rodata like you'd hope. Fix this with the usual string table trick.
Cuts about 20k (-80k read-write +60k read-only) and ~280 relocations
from the gallium driver.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11168>
