ppir_do_one_node_to_instr merges instructions and uses a pipeline reg
to save a reg. It tests if ppir_node_has_single_src_succ, but it should
check if ppir_node_has_single_succ.
The following deqp tests run into this issue because they have a node
with 2 successors in different blocks, where one was merged into the same
instruction and the second one is pointing to a missing predecessor then.
Fixes the following deqp tests:
dEQP-GLES2.functional.shaders.loops.do_while_dynamic_iterations.vector_counter_fragment
dEQP-GLES2.functional.shaders.loops.for_dynamic_iterations.vector_counter_fragment
dEQP-GLES2.functional.shaders.loops.while_dynamic_iterations.vector_counter_fragment
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6555>
For NIR-to-TGSI, we don't want to revectorize 64-bit ops that we split to
scalar beyond vec2 width. We even have some ops that we would rather
retain as scalar due to TGSI opcodes being scalar, or having more unusual
requirements.
This could be used to do the vectorize_vec2_16bit filtering, but that
shader compiler option is also used in algebraic so leave it in place for
now.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6567>
Incorrect implementation has been found during code surfing.
v3d implementation used for reference.
Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Reviewed-by: Icenowy Zheng <icenowy@aosc.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6076>
Instead of having separate lists of variables, roughly sorted by mode,
use a single list for all shader-level NIR variables. This makes a few
list walks a bit longer here and there but list walks aren't a very
common thing in NIR at all. On the other hand, it makes a lot of things
like validation, printing, etc. way simpler. Also, there are a number
of cases where we move variables from inputs/outputs to globals and this
makes it way easier because we no longer have to move them between
lists. We only have to deal with that if moving them from the shader to
a nir_function_impl.
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5966>
Gallium drivers should never see nir_var_uniform because gallium lowers
regular uniforms to a UBO. No GL driver should ever see either
nir_var_mem_shared because that's lowered in GLSL IR.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5418>
All squashed into a single commit because it shouldn't have any
behaviour change, except that it might work now on platforms where it
was broken because F_DUPFD_CLOEXEC is not supported but FD_CLOEXEC is.
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5369>
This allows us to do API specific checks before removing variable
without filling nir_remove_dead_variables() with API specific code.
In the following patches we will use this to support the removal
of dead uniforms in GLSL.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4797>
After the recent optimizations in ppir lowering that increase options
for combining, node_to_instr now may have multiple options of nodes to
insert and needs to decide which is better.
For example, if an instruction uses both a varying and a texture, there
are two nodes nodes that can be inserted to the load varying slot in the
same instruction (ld_var and ld_coords). It is much more advantageous to
pipeline the load texture coords since that enables the higher precision
path for texture coordinates. However, with the current recursive
expansion, this cannot be influenced.
This simple ready list implementation in node_to_instr allows it to
choose the next node to expand based on a priority score, rather than
relying on the random order coming from the recursive expansion.
Other than preferring nodes with pipeline output (which covers ld_coords
vs ld_var), nodes using later slots in the pipeline are now expanded
first, allowing node_to_instr to make all of the earlier (pipelineable)
nodes available in the ready list so the best one can be chosen when
picking nodes for the earlier slots.
Fixes: 632a921bd0 lima/ppir: optimize tex loads with single successor
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5092>
Move steps from lowering to emit, since they can be done earlier in a
single place, rather than in two-steps.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4975>
ppir doesn't register successors in other blocks, and causes
ppir_node_has_single_succ to be unreliable as it might return true for
nodes with successors in other blocks.
This is bad for optimization passes that try to pipeline registers or
avoid insertion of movs, as that can generally only be done for nodes
with a single user.
As of now, ppir can't just start adding successors in other blocks as
that breaks the scheduling code.
So this patch is a little hacky but enables pipelining optimizations
during lowering. It can hopefully be removed during future scheduler
rework.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4975>
In some paths where ppir_emit_cf_list is called, compilation errors such
as in unsupported features were not being handled, allowing compilation
to continue and fail at some random point later.
Handle them properly so compilation aborts in the expected way rather
than what may look like a compiler crash/bug.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4975>
For now this parameter doesn't do anything.
It means the implementation is allowed to use
a cache on disk.
Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4993>
This is yet another simple optimization that attemts to save the
insertion of an unnecessary mov for a large number of cases.
If the node outputting the condition for select satisfies a few
requirements (which are common in the case of comparison conditions),
it can just be changed to pipeline output and used directly.
In case of difficult corner cases, just fall back to the mov as before.
The sel_cond op is removed as the scheduler can be smart enough to place
nodes that output to ^fmul in the ALU_SCL_MUL slot, and as there can be
alu ops other than just mov.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4632>
It turns out that with more aggressive combining, there can be cases
where the available const slots are not enough for one instruction.
In particular, fcsel can take up to two consts, and a previous alu slot,
such as a comparison condition, might require an additional const.
So add a fallback for it like for uniforms.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4632>
In many cases, it is possible to avoid creating a mov for the store
output node.
Additionally, nodes other than alu, such as load varying, can be valid
store output nodes too.
This is another small optimization, but helps a vast majority of
programs by 1 instruction.
Shaders with discard easily become complicated to handle properly.
Some example issues: ppir has to rely on instruction ordering; or a
node with ssa output could be required both before a discard_if (as a
condition) and after it (as the instruction with the 'stop' bit set).
So don't try to handle them here.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4632>
The previous code assumed that a ppir node would be created for each nir
instr and used that to add it to the list of nodes and verify success.
This didn't make much sense anymore since some emit paths create
multiple nodes anyway, and this didn't allow for an emit call to not
create any new ppir node while still returning success.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4632>
With the previous refactors moving these lowering steps to a nir pass,
these are no longer needed.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
Move the duplicate consts step to a nir pass.
This makes the nir representation closer to what ppir will have in the
result.
Additionally, it handles the case where a const is used multiple times
by a single node (which can happen in instructions like fcsel). The new
implementation will only emit a single load const for that case.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
Move the duplicate uniform and varying steps to a nir pass, along with
some changes in the duplicating strategy.
Node duplication is now done per user of the varying/uniform. This is
inspired by what the offline shader compiler seems to usually do, and as
usual aims to reduce register pressure and better utilize the ld_uni and
ld_var instruction slots.
It is worth noting that due to a bug/feature, ppir was already
duplicating uniforms per successor in ppir_node_add_src even if the
comment indicated it was meant to be per-block.
Additionally, ppir was duplicating load uniform nodes twice for nodes
that use the same uniform in more than one source, resulting in one
unnecessary (and unpipelineable) load. This new implementation in nir
only creates one load in that case.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
Varying loads with a single successor have a high potential to be
combined with its successor node, like ppir does for uniforms, rather
than being in a separate instruction.
Even if ppir becomes capable of combining instructions in a separate
step, combining varying loads during node_to_instr is trivial enough
that it seems to be worth doing it in this stage, and this benefits
pretty much every program that uses varyings.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
Even if a node has pipeline output and a single successor, it is still
valid for that successor to have multiple references to that pipeline
node. A trivial example is add(u.x,u.y) where u is a uniform.
It is even possible for this to occur with consts as operands of fcsel.
So remove uses of ppir_node_get_src_for_pred as that would assume a
single src in the node that uses the pipeline.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
The lod bias register is correctly run through the entire compilation
process, but in the end its allocated register value was never being
added to the instruction.
It seems that most programs were lucky enough that lod bias was assigned
register 0.x so that things worked anyway.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
The current solution for handling registers that live and die within a
single instruction does not handle all cases. In particular, these
intra-instruction use register also conflict with registers that are
part of the live_in set.
Unfortunately, adding them to the live_in set is not an easy solution as
that would cause them to be propagated upwards. So, add a separate set
to handle these registers in the particular instructions, without
propagating them.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4535>
this adds a new pipe cap that drivers can support which enables passing buffer
clears with scissor test enabled through to be handled by the driver instead
of having mesa draw a quad
also adjust all existing clear() hooks to have the new parameter
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4310>
With EGL_EXT_image_dma_buf_import, user can import dma_buf
with offset.
This is also used by AOSP GLConsumer::updateTexImage
with HAL_PIXEL_FORMAT_YV12 buffer which store YUV planes in
the same buffer with offset. Render sample from it using
GL_OES_EGL_image_external. This should fix some video
display problem when using MediaCodec soft decoding which
generates HAL_PIXEL_FORMAT_YV12 buffer and render it on
screen.
Test program:
https://github.com/yuq/gfx/tree/master/yuv2rgb/dma-buf
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4362>
This is preparation for the next commit where we may need different
swap_r_b flags for pixel and texel formats.
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4241>
Fragment shader can write depth and stencil if we set necessary flags
in RSW. In addition to that we need to use special format for Z24S8.
Original format is apparently Z24X8 since we can't sample stencil in GLES2.
This new format also seems to use several components for storing depth
since we saw r != g != b when sampling with this format.
[vasily: - initialize clear->depth to 0xffffff if we reload depth, just
like blob does. Reloading doesn't work otherwise
- use single bitmap for reload type]
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4197>
Unfortunately we don't know how to reload Z16 buffers yet and blob
is using Z24 for dEQP tests that need depth reload.
Reviewed-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4197>
GP doesn't support ftrunc natively and unfortunately one in generic
opt_algebraic is not GP-friendly either. Introduce our own lowering
that utilizes fsign() that GP supports:
ftrunc(a) = fmul(fsign(a), ffloor(fmax(a, -a)))
Tested-by: Andreas Baierl <ichgeh@imkreisrum.de>
Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4126>
We also add a DCE pass to cleanup the result of this pass, which turns
out to also be necessary to cleanup the result of nir->gpir in some
cases that we didn't hit until the next commit.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4125>
Optimize the result of a conditional break/continue. In NIR something
like:
loop {
...
if (cond)
continue;
would get lowered to:
block_0:
...
block_1:
branch_cond !cond block_3
block_2:
branch_uncond block_0
block_3:
...
We recognize the conditional branch skipping over the unconditional
branch, and turn it into:
block_0:
...
block_1:
branch_cond cond block_0
block_2:
block_3:
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4125>
We weren't using this function before. The name is confusing, but it
changes the child while also fixing up the dependence link, if you don't
have access to it already. Or at least, I think that's what the
intention is, and what we'll need to change the branch condition in the
next commit. Adding a dependency between the new and old source doesn't
make any sense for this, and we also need to change the actual source.
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4125>
