Copy propagation often eliminates all uses of an instruction. If we
detect that we've done so, we can eliminate the instruction ourselves
rather than leaving it hanging until the next DCE pass.
This saves some CPU time as other passes don't see dead code.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
The new def-based pass works better in many cases, and should be less
resource intensive. However, the limited visibility of the defs-based
pass due to many values not being SSA yet makes it unable to fully
replace the old pass. Try the new one, and if it can't make progress,
then try the old one. That way, things will mostly be handled by the
new pass, but everything that was being cleaned up still will be.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
While the limited visibility due to partial SSA is a downside to the new
pass, it has a huge number of advantages that make it worth switching
over even now. It's much more efficient, can eliminate redundant memory
loads across blocks, and doesn't generate loads of unnecessary copies
that other passes have to clean up. This means we also eliminate the
infighting between the old CSE, coalescing, and copy propagation passes.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
This has a number of advantages compared to the pass I wrote years ago:
- It can easily perform either Global CSE or block-local CSE, without
needing to roll any dataflow analysis, thanks to SSA def analysis.
This global CSE is able to detect and coalesce memory loads across
blocks. Although it may increase spilling a little, the reduction
in memory loads seems to more than compensate.
- Because SSA guarantees that values are never written more than once,
the new CSE pass can directly reuse an existing value. The old pass
emitted copies at the point where it discovered a value because it
had no idea whether it'd be mutated later. This led it to generate
a ton of trash for copy propagation to clean up later, and also a
nasty fragility where CSE, register coalescing, and copy propagation
could all fight one another by generating and cleaning up copies,
leading to infinite optimization loops unless we were really careful.
Generating less trash improves our CPU efficiency.
- It uses hash tables like nir_instr_set and nir_opt_cse, instead of
linearly walking lists and comparing each element. This is much more
CPU efficient.
- It doesn't use liveness analysis, which is one of the most expensive
analysis passes that we have. Def analysis is cheaper.
In addition to CSE'ing SSA values, we continue to handle flag writes,
as this is a huge source of CSE'able values. These remain block local.
However, we can simply track the last flag write, rather than creating
entire sets of instruction entries like the old pass. Much simpler.
The only real downside to this pass is that, because the backend is
currently only partially SSA, it has limited visibility and isn't able
to see all values. However, the results appear to be good enough that
the new pass can effectively replace the old pass in almost all cases.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
Like NIR, we print SSA defs as %1, %2, and so on. The number here is
the VGRF number. VGRFs that don't correspond to a SSA def remain
printed as vgrf1, vgrf2, and so on.
This makes it much easier to see what values are SSA and which aren't.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
Even without a full use list, simply tracking the number of uses will
let us tell "this is the only use of the def" or "we've just replaced
all uses of a def". It's inexpensive to calculate and will be useful.
(rebased by Kenneth Graunke)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
This introduces a new analysis pass that opportunistically looks for
VGRFs which happen to satisfy the SSA definition properties.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
Our code to initialize gl_SubgroupInvocation uses multiple instructions
some of which are partial writes. This makes it difficult to analyze
expressions involving gl_SubgroupInvocation, which appear very
frequently in compute shaders.
To make this easier, we add a new virtual opcode which initializes
a full VGRF to the value of gl_SubgroupInvocation. (We also expand
it to UD for SIMD8 so there are not partial write issues.) We then
lower it to the original code later on in compilation, after we've
done the bulk of our optimizations.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
This gathers a number of sources into a contiguous vector register,
typically using LOAD_PAYLOAD. However, it uses MOV for a single source.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28666>
These allow avoiding dead-locks in non-compliant applications that
execute barriers under non-uniform control flow. They're not expected
to have any major disadvantage so let's enable them unconditionally.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29562>
See also HSDES#14015504893 regarding the region-based tessellation
redistribution feature which allows fine-tuning the number of regions
per patch. This sets it to the recommended value, since region-based
redistribution is enabled by default.
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29562>
This flag is mostly redundant with uses_discard and was only
introduced to implement demote with LLVM when it didn't have
that intrinsic.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27617>
The semantics of discard differ between GLSL and HLSL and
their various implementations. Subsequently, numerous application
bugs occurred and SPV_EXT_demote_to_helper_invocation was written
in order to clarify the behavior. In NIR, we now have 3 different
intrinsics for 2 things, and while demote and terminate have clear
semantics, discard still doesn't and can mean either of the two.
This patch entirely removes nir_intrinsic_discard and
nir_intrinsic_discard_if and replaces all occurences either with
nir_intrinsic_terminate{_if} or nir_intrinsic_demote{_if} in the
case that the NIR option 'discard_is_demote' is being set.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27617>
Previously, we left it NULL until later in the compile. However, some
builder helpers are starting to check the options and they blow up when
options == NULL.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27617>
The MCS region to ambiguate needs to shift 4KB from its
starting address. The first 4KB is reserved for hardware.
Signed-off-by: Jianxun Zhang <jianxun.zhang@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28919>
The scaledown rectangle of MSAA fast clear on Xe2 is 8 times
in X and 2 in Y dimension of previous platforms.
Absorb refactoring change suggested by
Nanley Chery <nanley.g.chery@intel.com>
Signed-off-by: Jianxun Zhang <jianxun.zhang@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28919>
MOV_INDIRECT picks one lane from the src[0] and moves it to all lanes
in the destination. Even if we split the instruction, src[0] should
remain identical.
Noticed this while trying to use this instruction in SIMD32. All
current use cases are limited to SIMD8 shaders (or SIMD16 on Xe2). Or
maybe in SIMD32 but with a uniform src[0]. That's we think we've never
seen the issue so far.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28036>
Since we don't need to share that data with other fixed functions.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29571>
All the MI_SDI currently have forced write checks (meaning the command
streamer will stall until completion) on Gfx12.0+.
Now on Gfx12.0/12.5, the read commands have implicit waits on previous
writes (BSpec ). So if we're only dealing with CS writes & reads, we
don't need forced write checks.
In the few cases where CS is writing data for other bits of HW, we
need the forced write checks. This change adds an API that will let
the driver decide when to enable forced write checks.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29571>
When you want to write a value to a register or memory but you don't
know just yet that value when you emit the command.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29571>
With this change, the engine initialization batches are build and
submitted at vkCreateDevice() but the function doesn't wait for them
to complete. Instead we wait at vkDestroyDevice() or whenever another
submission happens on the queue, we check whether the initialization
batch has completed (without waiting) and free it if completed.
Seems to be about 25% reduction time of vkCreateDevice()
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28975>
The array pool does a single allocation and then splits it out. The
downside is that the pool is not lockless, but for border colors it
likely doesn't matter much as there is a max border colors for 4k.
Seems to be a 30% time reduction for vkCreateDevice()
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28975>
We can remove a bunch of TRTT specific code from the backends as well
as manual submission tracking.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28975>
We want to make this more generic so that it can be reused for device
initialization as well as TRTT submissions.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28975>
Drop usage of pthread mutex so initialization never fails.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28975>
According to PRMs:
"All parameters are of type IEEE_Float, except those in the The ld*,
resinfo, and the offu, offv of the gather4_po[_c] instruction message
types, which are of type signed integer."
Currently, we load parameters with the correct types, but use them as send
sources with the default float type, which may confuse passes downstream.
Fix this by actually storing the retyped sources.
Cc: mesa-stable
Signed-off-by: Sviatoslav Peleshko <sviatoslav.peleshko@globallogic.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29581>
According to PRMs:
"All parameters are of type IEEE_Float, except those in the The ld*,
resinfo, and the offu, offv of the gather4_po[_c] instruction message
types, which are of type signed integer."
Currently, we load parameters with the correct types, but use them as send
sources with the default float type, which may confuse passes downstream.
Fix this by actually storing the retyped sources.
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11118
Signed-off-by: Sviatoslav Peleshko <sviatoslav.peleshko@globallogic.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29581>
This AUX-TT is only updated on the CPU since ee6e2bc4a3 ("anv: Place
images into the aux-map when safe to do so"). So the only really
important invalidation that needs to happens is on the beginning of a
primary command buffer.
We are required to idle the pipes prior invalidation the AUX-TT. This
might not be happening when the invalidation is put at the beginning
of the secondary command buffers.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29671>
