Use anv_state_pool_return_blocks() to return blocks to the pool, instead
of manually pushing them.
v3:
- return blocks from the end of the chunk (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The use of anv_state_table_add() combined with anv_state_table_push(),
specially when adding a bunch of states to the table, is very verbose.
So we add this helper that makes things easier to digest.
We also already add the anv_state_table member in this commit, so things
can compile properly, even though it's not used.
v2: assert that the states are always aligned to their size (Jason)
v3: Add "table" member to anv_state_pool in this commit.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We will need the anv_block_pool_map to find the map relative to some BO
that is not at the start of the block pool.
v2: just return a pointer instead of a struct (Jason)
v4: Update comment (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Add a structure to hold anv_states. This table will initially be used to
recycle anv_states, instead of relying on a linked list implemented in
GPU memory. Later it could be used so that all anv_states just point to
the content of this struct, instead of making copies of anv_states
everywhere.
One has to call anv_state_table_add(), which returns an index for the
state in the table, and then get a pointer to such index, and finally
fill in the rest of the struct.
TODO:
1) There's a lot of common code between this table backing store
memory and the anv_block_pool buffer, due to how we grow it. I think
it's possible to refactory this and reuse code on both places.
2) Add unit tests.
v3:
- Rename state table memfd (Jason)
- Return VK_ERROR_OUT_OF_HOST_MEMORY on more places (Jason)
- anv_state_table_grow returns VkResult (Jason)
- Rename variables to be more informative (Jason)
- Return errors on state table grow.
- Rename anv_state_table_push/pop to anv_free_list_push2/pop2
This will be renamed again to remove the trailing "2" later.
v4:
- Remove exit(-1) from anv_state_table (Jason).
- Use uint32_t "next" field in anv_free_entry (Jason).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
There were 2 problems with this test.
First it was comparing highest, which was -1, with an uint32_t. So the
current value would never be higher than that, and the assert would
always be false. It just never reached this point because of the next
problem.
It was always looking for the highest value of each thread and storing
it in thread_max. So a test case like this wouldn't work:
[Thread]: [Blocks]
[0]: [0, 32, 64, 96]
[1]: [128, 160, 192, 224]
[2]: [256, 288, 320, 352]
Not only that would skip values and iterate only over thread number 2,
instead of walking through all of them, but thread_max was also
initialized to -1. And then compared to unsigned blocks[i][next[i].
We fix that by getting the smallest value of each thread, and checking
if it is lower than thread_min, which is initialized to INT32_MAX. And
then we end up walking through all the blocks of all threads. We also
change "blocks" to be int32_t instead of uint32_t, since in some places
(alloc_blocks) it was already referenced as int32_t, and that fixes the
comparison to -1.
v2:
- keep highest initialized to -1, and change blocks to be int32_t.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The ++ operator strikes again.
Fixes: f92c5bc8f3 ("anv/device: fix maximum number of images supported")
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We had defined MAX_IMAGES as 8, which we used to size the array for
image push constant data. The comment there stated that this was for
gen8, but anv_nir_apply_pipeline_layout runs for all gens and writes
that array, asserting that we don't exceed that number of images,
which imposes a limit of MAX_IMAGES on all gens.
Furthermore, despite this, we are exposing up to 64 images per shader
stage on all gens, gen8 included.
This patch lowers the number of images we expose in gen8 to 8 and
keeps 64 images for gen9+ while making sure that only pre-SKL gens
use push constant space to handle images.
v2:
- <= instead of < in the assert (Eric, Lionel)
- Change the way the assertion is written (Eric)
v3:
- Revert the way the assertion is written to the form it had in v1,
the version in v2 was not equivalent and was incorrect. (Lionel)
v4:
- gen9+ doesn't need push constants for images at all (Jason)
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> (v3)
Fixes following failing vk-gl-cts cases on Linux desktop:
dEQP-VK.api.external.memory.android_hardware_buffer.suballocated.buffer.info
dEQP-VK.api.external.memory.android_hardware_buffer.suballocated.image.info
dEQP-VK.api.external.memory.android_hardware_buffer.dedicated.image.info
dEQP-VK.api.external.memory.android_hardware_buffer.dedicated.buffer.info
Fixes: 517103abf1 "anv/android: add ahardwarebuffer external memory properties"
Reported-by: Juan A. Suarez <jasuarez@igalia.com>
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Replace calls to create hash tables and sets that use
_mesa_hash_pointer/_mesa_key_pointer_equal with the helpers
_mesa_pointer_hash_table_create() and _mesa_pointer_set_create().
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Acked-by: Eric Engestrom <eric@engestrom.ch>
This function is modeled after the aux_op functions except that it has a
lot more parameters because it deals with two images as well as source
and destination regions.
The original idea was that the backend compiler could eliminate
surfaces, so we would have it mark which ones are actually used,
then shrink the binding table accordingly. Unfortunately, it's a
pretty blunt mechanism - it can only prune things from the end,
not the middle - since we decide the layout before we even start
the backend compiler, and only limit the size. It also basically
gives up if it sees indirect array access.
Besides, we do the vast majority of our surface elimination in NIR
anyway, not the backend - and I don't see that trend changing any
time soon. Vulkan abandoned this plan a long time ago, and I don't
use it in Iris, but it's still been kicking around in i965.
I hacked shader-db to print the binding table size in bytes, and
observed no changes with this patch. So, this code appears to do
nothing useful.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
The num_components value passed into get_mul_for_src is used to only
compose the parts of the swizzle that we know will be used so we don't
compose invalid swizzle components. However, we had a bug where we
passed the number of components of the add all the way through. For the
given source, we need the number of components read from that source.
In the case where we have a narrow add, say 2 components, that is
sourced from a chain of wider instructions, we may not compose all the
swizzles. All we really need to do is pass through the right number of
components at each level.
Fixes: 2231cf0ba3 "nir: Fix output swizzle in get_mul_for_src"
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This will allow drivers to pin shader buffers if necessary.
i965 and anv do not need to do this today, but iris will.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Currently, BLORP expects drivers to provide two functions for dealing
with buffers: blorp_emit_reloc and blorp_surface_reloc. Both record a
relocation and combine the BO address and offset into a full 64-bit
address. Traditionally, blorp_surface_reloc has written that combined
address to an implicitly-known buffer where surface states are stored.
(In contrast, blorp_emit_reloc returns the value.)
The upcoming Iris driver stores surface states in multiple buffers,
which makes it impossible for blorp_surface_reloc to write the combined
address - it only takes an offset, not the actual buffer to write to.
This commit adds a third function, blorp_get_surface_address, which
combines and returns an address, which is then passed to ISL's surface
state fill functions. Softpin-only drivers can return a real address
here and skip writing it in blorp_surface_reloc. Relocation-based
drivers are have options. They can simply return 0 from the new
function, and continue writing the address from blorp_surface_reloc.
Or, they can return a presumed address from blorp_get_surface_address,
and have other relocation processing write the real value later.
For now, i965 and anv simply return 0.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This adds a second level of caching for the pre-lowered NIR that's only
based off of the shader module, entrypoint and specialization constants.
This is enough for spirv_to_nir as well as our first round of lowering
and optimization. Caching at this level should allow for faster shader
recompiles due to state changes.
The NIR caching does not get serialized to disk via either the
VkPipelineCache serialization mechanism or the transparent on-disk
cache. We could but it's usually not that expensive to fall back to
SPIR-V for the odd cache miss especially if it only happens once for
several misses and it simplifies the cache.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
The stuff hashed by anv_pipeline_hash_shader is exactly the inputs to
anv_shader_compile_to_nir so it can be used for NIR caching.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Thanks to the new NIR load_descriptor intrinsic added by the UBO/SSBO
lowering series, we weren't getting UBO pushing because the UBO range
detection pass couldn't see the constants it needed. This fixes that
problem with a quick round of constant folding. Because we're folding
we no longer need to go out of our way to generate constants when we
lower the vulkan_resource_index intrinsic and we can make it a bit
simpler.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Patch moves intel_tiled_memcpy[_sse41] libraries to isl, renames some
functions and types and makes the required build system changes for
meson, automake and Android. No functional changes are introduced.
v2: code cleanups, move isl_get_memcpy_type to i965 (Jason)
v3: move isl_mem_copy_fn to priv header, cleanups (Jason, Dylan)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Shaders containing software implementations of double-precision
operations can be very large such that we cannot stack-allocate
an array of grf_count*16.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Shaders containing software implementations of double-precision
operations can be very large such that we have more the 2^16 virtual
registers during optimization.
Move the 'nr' field to the union containing the immediate storage and
expand it to 32-bits.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The next patch replaces an unsigned bitfield with a plain unsigned,
which triggers gcc to begin warning on signed/unsigned comparisons.
Keeping this patch separate from the actual move allows bisectablity and
generates no additional warnings temporarily.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
A follow on commit will move nr to the same union as the immediate
data, so we should assert these invariants before we overwrite the nr
field.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
A follow on patch will move the 'nr' field to the union containing the
immediate field, so prepare by checking that we're only testing these
assertions if the .file is correct.
The assertions with != ARF were kind of silly to begin with because the
<128 check is specifically only for things in the GRF.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
These are broken on a future platform, but it turns out we don't need
to fix them, since they're just type-converting moves with strided
source. Kill them.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This legalization pass is meant to handle situations where the source
or destination regioning controls of an instruction are unsupported by
the hardware and need to be lowered away into separate instructions.
This should be more reliable and future-proof than the current
approach of handling CHV/BXT restrictions manually all over the
visitor. The same mechanism is leveraged to lower unsupported type
conversions easily, which obsoletes the lower_conversions pass.
v2: Give conditional modifiers the same treatment as predicates for
SEL instructions in lower_dst_modifiers() (Iago). Special-case a
couple of other instructions with inconsistent conditional mod
semantics in lower_dst_modifiers() (Curro).
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Currently the visitor attempts to enforce the regioning restrictions
that apply to double-precision instructions on CHV/BXT at NIR-to-i965
translation time. It is possible though for the copy propagation pass
to violate this restriction if a strided move is propagated into one
of the affected instructions. I've only reproduced this issue on a
future platform but it could affect CHV/BXT too under the right
conditions.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
I triggered this bug while prototyping code for a future platform on
IVB. Could be a problem today though if a strided move is
copy-propagated into a type-converting move with DF destination.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This seems to be a problem in combination with the lower_regioning
pass introduced by a future commit, which can modify a SIMD-split
instruction causing its execution size to become illegal again. A
subsequent call to lower_simd_width() would hit this bug on a future
platform.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Align16 is no longer a thing, so a new implementation is provided
using Align1 instead. Not all possible swizzles can be represented as
a single Align1 region, but some fast paths are provided for
frequently used swizzles that can be represented efficiently in Align1
mode.
Fixes ~90 subgroup quad swap Vulkan CTS tests.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
lower_integer_multiplication() implements 32x32-bit multiplication on
some platforms by bit-casting one of the 32-bit sources into two
16-bit unsigned integer portions. This can give incorrect results if
the original instruction specified a source modifier. Fix it by
emitting an additional MOV instruction implementing the source
modifiers where necessary.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Looks like it is impossible that 'last' variable is a null
because at least the get_vs_prog_data shouldn't return a null pointer.
So this check is unnecessary starts from commit:
99d497c5b6 "anv/pipeline: Replace get_fs_input_map with ..."
This small issue is found by cppcheck.
Signed-off-by: Andrii Simiklit <andrii.simiklit@globallogic.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
