This consolidates several duplicated pieces of code into devinfo.
max_scratch_ids is an array that provides the max number of threads
for the rendering and compute stages.
This fixes some exceptions missed by crocus for scratch ids on haswell
and cherryview.
It also fills out devinfo->max_scratch_ids properly for stages VS
through CS on Gfx12.5. But, functionally this should not make a
difference as Gfx12.5 already uses COMPUTE for all stages.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12799>
iris, i965, and anv define the PAGE_SIZE in anv_allocator and bufmgr
files. As on FreeBSD the page size is defined in machine/param.h that is
indirectly included by those files, we'd rather define it only when the
system is not FreeBSD to avoid compile errors.
v2: Changed the path in the comment to make clear that machine/params.h
is a FreeBSD system file.
Signed-off-by: Eleni Maria Stea <elene.mst@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11203>
This is the first time we see an application running out of mmap().
We essentially allocate too many batches (+65k) and end up not being
able to mmap them, at which point we can't mmap anything anymore and
things go sideways.
This change allocates bigger batch BOs as we grow an existing command
buffer. This drastically reduces the number of BOs we need to allocate
(the benchmark that reported the issue now reaches a max of ~630 BOs,
instead of reaching 65k and failing previously).
v2: Track the total batch size of command buffers (Jason)
Just give 0 for batch_len to i915 (Jason)
v3: Fix indentation (Jason)
v4: Drop uncessary reshuffling of error labels (Jason)
v5: Remove empty lines (Marcin)
v6: Limit BO growing to chunks of 16Mb (Jason)
v7: Add assert on initial size (Jason)
v8: Add define for max size (Jason)
v9: Fixup v7 assert for non softpin platforms (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4956
Cc: mesa-stable
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11482>
Rework:
* Jordan: Handle per_thread_scratch==0 in anv_scratch_pool_get_surf
* Jordan: Update subslices in anv_scratch_pool_alloc
* Jason: Clean up the patch a bit
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11582>
This makes a bunch of loops use ARRAY_SIZE instead of MESA_SHADER_STAGES,
extends a few arrays, and adds a bunch of array length asserts.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8637>
This patch renames all macros with "GEN_" prefix defined in
common code.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9413>
This patch renames functions, structures, enums etc. with "gen_"
prefix defined in common code.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9413>
Changes in this patch include:
- Rename all files in src/intel/common path
- Update the filenames used in source and build files
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9413>
This is way better than the giant macro thing we had before.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8676>
According to the BSpec page for MEDIA_VFE_STATE, on Gen12 platforms
"if a fused configuration has fewer threads than the native POR
configuration, the scratch space allocation is based on the number of
threads in the base native POR configuration". However we currently
use the subslice count from devinfo->num_subslices[0], which only
includes the subslices currently enabled by the platform fusing. This
leads to scratch space underallocation and occasional hangs.
The problem is likely to affect most Gen12 GPUs with less than 96 EUs.
GFXBench5 Aztec Ruins is able to reproduce the issue fairly reliably.
Fixes: 9e5ce30da7 "intel: fix the gen 12 compute shader scratch IDs"
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8636>
In d11e4738a8, we started using a start_offset to allow us to
allocate pools where the base address isn't at the start of the pool.
This is useful for binding table pools which want to be relative to
surface state base address (more or less), among other things. However,
we had a bug where, if you have a negative offset, everything returned
to the pool would end up being returned to the "back" of the pool. This
isn't what we want for binding tables in the softpin world. This was
causing us to never actually re-use any binding table blocks. How this
passed CTS, I have no idea.
Closes: #3100
Fixes: d11e4738a8 "anv/allocator: Add a start_offset to anv_state_pool"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5395>
This allows a pool's allocations to start somewhere other than the base
address. Our first real use of this will be to use a negative offset
for the binding table pool to make it so that the offset is baked into
the pool and the code in anv_batch_chain.c doesn't have to understand
pool offsetting.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4897>
Also update all of its callers.
On the next commit, the device will be used by anv_gem_munmap to choose
whether we need to call the valgrind code or not, depending on which
type of mmap we are using.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1675>
When we originally wrote a bunch of the allocation data structures, we
re-used the GPU memory for CPU-side data structures. It's a bit more
memory efficient and usually ok. However, this has a couple of
problems:
1. It makes it MUCH more likely that the GPU will accidentlly stomp
CPU-side data structures and cause nearly impossible to debug
crashes.
2. With discrete GPUs, the memory will be mapped somehow and that map
may be across the BAR so it could have horribly slow CPU access.
This is bad for our CPU-side data structures.
In the case of anv_state_stream, it also made the data structure
massively more complex than it needed to be.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>
If we have an allocation that's exactly the block size, we end up
computing a new block size to allocate that's exactly the block size,
add in the header, and then assert fail. When computing the block size,
we need to account for the header.
Fixes: 955127db93 "anv/allocator: Add support for large stream..."
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>
My understanding is that there's no reason for the scratch space
allocation to be different between iris, i965 and anv. Let's make all
the functions behave the same.
I don't know if this fixes any specific gen9 bugs, it it might since
it increases the scratch space.
v2: Rebase.
v3: Rebase.
v4: Remove redundant gen 11 check (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4006>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4006>
This is the same idea as "intel: fix the gen 11 compute shader scratch
IDs".
The number of EUs on TGL is not the same as ICL, but the
MEDIA_VFE_STATE restrictions stay the same, so adapt the code to it.
Also, consider the base configuration instead of what we read from the
Kernel.
According to Mark, this fixes the following piglit tests on TGL:
piglit.spec.arb_compute_shader.execution.shared-atomicmax-uint.tglm64
piglit.spec.arb_compute_shader.execution.shared-atomicmax-int.tglm64
piglit.spec.intel_shader_atomic_float_minmax.execution.shared-atomicmax-float.tglm64
v2: s/ICL+/Gen11+/ (Jason).
Cc: mesa-stable@lists.freedesktop.org
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4006>
Scratch space allocation is based on the number of threads in the base
configuration, and we only have one base configuration for ICL, with 8
subslices.
This fixes an issue with Aztec on Vulkan in a machine with a
configuration that's not the base. The issue looks like a regression
from b9e93db208, but it seems things are broken since forever, just
not easily reproducible.
v2: Reimplement it using the subslices variable. Don't touch TGL.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4006>
Because softpin block pools are made up of a set of BOs with different
maps, it was possible for a single state to end up straddling blocks.
To fix this, we pass a contiguous size to anv_block_pool_grow and it
ensures that the next allocation in the pool will have at least that
size.
We also add an assert in anv_block_pool_map to ensure we always get
contiguous maps. Prior to the changes to anv_block_pool_grow, the unit
tests failed with this assert. With this patch, the tests pass.
This was causing problems on Gen12 where we allocate the pages for the
AUX table from the dynamic state pool. The first chunk, which gets
allocated very early in the pool's history, is 1MB which was enough that
it was getting multiple BOs. This caused the gen_aux_map code to write
outside of the map and overwrite the instruction state pool buffer which
lead to GPU hangs.
Fixes: 731c4adcf9 "anv/allocator: Add support for non-userptr"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The previous way we were attempting to handle AUX tables on TGL-LP was
very GL-like. We used the same aux table management code that's shared
with iris and we updated the table on image create/destroy. The problem
with this is that Vulkan allows multiple VkImage objects to be bound to
the same memory location simultaneously and the app can ping-pong back
and forth between them in the same command buffer. Because the AUX
table contains format-specific data, we cannot support this ping-pong
behavior with only CPU updates of the AUX table.
The new mechanism switches things around a bit and instead makes the aux
data part of the BO. At BO creation time, a bit of space is appended to
the end of the BO for AUX data and the AUX table is updated in bulk for
the entire BO. The problem here, of course, is that we can't insert the
format-specific data into the AUX table at BO create time.
Fortunately, Vulkan has a requirement that every TILING_OPTIMAL image
must be initialized prior to use by transitioning the image from
VK_IMAGE_LAYOUT_UNDEFINED to something else. When doing the above
described ping-pong behavior, the app has to do such an initialization
transition every time it corrupts the underlying memory of the VkImage
by using it as something else. We can hook into this initialization and
use it to update the AUX-TT entries from the command streamer. This way
the AUX table gets its format information, apps get aliasing support,
and everyone is happy.
One side-effect of this is that we disallow CCS on shared buffers.
We'll need to fix this for modifiers on the scanout path but that's a
task for another patch. We should be able to do it with dedicated
allocations.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3519>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3519>
All they do now is take a size, align, and flags and figure out which
heap to allocate in. All of the actual code to deal with the BO is in
anv_allocator.c. We want to leave anv_vma_alloc/free in anv_device.c
because it deals with API-exposed heaps so it still makes sense to have
it there.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3519>
Having to always pull the physical device from the instance has been
annoying for almost as long as the driver has existed. It also won't
work in a world where we ever have more than one physical device. This
commit adds a new field called "physical" to anv_device and switches
every location where we use device->instance->physicalDevice to use the
new field instead.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3461>
When we moved from allocating BOs directly to using the BO cache, we
lost the EXEC_OBJECT_CAPTURE flag on all our state buffers.
Fixes: 3119b96bdf "anv: Allocate block pool BOs from the cache"
Fixes: ee77938733 "anv: Allocate batch and fence buffers from..."
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
When a BO is flagged as having a client visible address, we put it in
its own heap. We also support the client explicitly specifying an
address in said heap. If an address collision happens, we return false
from anv_vma_alloc which turns into a VK_ERROR_OUT_OF_DEVICE_MEMORY.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
We already have a mechanism for specifying that we want a fixed address
provided by the driver internals. We're about to let the client start
specifying addresses in some very special scenarios as well so we want
to pass this through to the allocation function.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
In ee77938733, we started using the BO cache for anv_bo_pool and
stopped using the bo_flags parameter. However, we never dropped it from
the struct or the init function.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
BOs are now only ever allocated through the BO cache so there's no need
to have these exposed.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
While we're here, we get rid of the locking and use a lock-free
algorithm. The chances of spilling contention are low and this is
actually a bit simpler in some ways.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This commit switches block pools over to being allocated from the BO
cache rather than being allocated manually by the block pool.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
All block pools are allocated with the same flags. There's no good
reason why it needs to be configurable.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This makes a number of changes to the current API:
1. Everything is renamed to anv_device_* instead of anv_bo_cache_*
because the BO cache is soon going to be the sole BO allocation path
and not some special case to make import/export work.
2. Drop the cache parameter. It's totally redundant with the device
and just annoying to keep typing.
3. Rework flags so that they go the convenient direction for usage in
ANV rather than whichever awkward way the i915 specified it to
maintain backwards compatibility. This also gives us the
opportunity to set some defaults.
4. Add flags for mapping and coherency.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
