Latency is estimanted duration of a single wave, ignoring others in the
CU. It is similar to the old cycles statistic except it it's more accurate
and considers memory operations.
The InvThroughput statistic is a combination of MaxWaves, Latency and the
portion of the wave's execution which does not use various resources.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8994>
This allows them to be scheduled properly and simplifies the assembler a
little.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8994>
Keep track of the current loop depth in Program and set the depth inside
Program::insert_block() instead of repeating it every time we insert one.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8994>
The texture and sampler descriptors are well separated in Vulkan,
let's add a new field to allow mixing sampler and texture descs.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9517>
On Bifrost the vertex/instance ID are preloaded in special registers,
no need to add special attribute entries.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9517>
There's a minus(1) modifier on the entries field. Take it into account.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9517>
UBO index assignment is a bit special in Vulkan, it's based on the
descriptor set layout, which doesn't know about shaders' internal UBOs
(our sysval UBOs). Extend the backend compilers so we can place sysval
UBOs where we want: after all explicit UBOs.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9517>
I'm just too lazy to implement the logic to prepare push constant
buffers in the Vulkan driver. Besides, Vulkan has explicit push
constants, which AFAIK is not handled in the compiler backends yet,
and that will probably conflict with the UBO -> push constant
promotion.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9517>
Currently the screen destruction closes the dup'ed fd, but not the
original renderonly gpu fd, which is kept around for the lifetime of
the renderonly.
Squashed revert of "vc4: Don't leak the GPU fd for renderonly usage."
(commit 99ef66c325) as requested by Eric.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6983>
The renderonly object is something the winsys creates, so the pipe
driver has no business in memcpying or freeing it. Move those bits
to the winsys.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6983>
Some opcodes come with both type and size variants. Right now, only the
size is taken into account. Extend the builder to provide wrappers that
take a nir_type in addition to the bitsize.
While at it, fix wrappers taking a compare operator to use the proper
.{i,s,u} variant based on the comparison (equal and non-equal should
use .i, other comparisons should use .{u,s}).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
We will have 2 compute jobs per indexed indirect draw, one doing the
min-max index search and one patching the cmdstream. The second compute
job needs to depend on the first one, as well as the previous indirect
draw job to avoid corrupting the indirect draw context which is shared
at the batch level (global dependency).
Instead of handling that case in panfrost_add_job(), extend
panfrost_add_job() to accept an explicit global dependency.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
This is needed for indirect draws so the compute job can patch the
vertex/tiler jobs which are following in the chain.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
A block that has all its successors empty is not necessarily a leaf
block in the CFG, and removing the JUMP in that causes the shader
to continue executing code from another block instead of exiting.
This reverts commit a496b41d50.
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
The old values seemed to work fine, but the ISA docs recommend 0x0,0x3,0xc
and 0xf:
COMPR==1: export half-dword enable. Valid values are: 0x0,3,c,f
[0] enables VSRC0 : R,G from one VGPR (R in low bits, G high)
[2] enables VSRC1 : B,A from one VGPR (B in low bits, A high)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9459>
Obviously this didn't affect correctness. Not sure about performance.
It also changes enabled_channels to match radeonsi.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: f29c81f863 ("aco: use VOP2 for v_cvt_pkrtz_f16_f32 if possible")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9459>
This allows skipping mutex locking. Don't take the aux context into account.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9356>
Currently only initialized for a6xx, mostly because that is the easiest
setup for me to test and debug at the moment. But the couple a6xx changes
should not require counterparts in older gens.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
Previously we were expecting cb0 to be user_buffer. (We did in some
cases upload it to a gpu buffer, but this was an internally allocated
buffer and not something subject to rebind.) But with TC it becomes
a gpu buffer.
(Technically, with pctx->const_uploader, we shouldn't hit the rebind
path for cb0, but better to not try to be overly clever.. sooner or
later that would bite us.)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
With threaded_context, CSO creation happens in the frontend thread,
which means it is no longer safe to do blits (if needed, for sampler
views with format that cannot be UBWC). So move this to the first
time that the sampler view is bound.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
With threaded_context, in the TC_TRANSFER_MAP_UNSYNC case, we are
getting called from the frontend thread, rather than driver thread.
So we need a different slab_child_pool for that.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
This will be used by threaded_context to avoid stalls in the
DISCARD_WHOLE_RESOURCE case (and DISCARD_RANGE cases that can
be promoted to DISCARD_WHOLE_RESOURCE).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
Split out the usage simplification from main part of transfer_map and
handle the threaded-context specific TC_TRANSFER_x flags.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
For threaded_context, to properly handle replace_buffer_storage, we'll
need to handle multiple "iterations" of a resource using the same
tracking in order to implement transfer_map() correctly.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
We didn't see this before without threaded_context because we (normally)
wouldn't upload cb0 (the slot u_blitter uses). But with cb0 getting
uploaded we could hit a leak due to constant state only being restored
in the fd_blitter_clear() path. Move cb0 save to the one path that uses
it.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
On android, this will show up in logcat, rather than being lost into the
ether.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
Useful for drivers to add some sanity checks to avoid/detect threading
issues caused by things that might be called (indirectly) from frontend
thread.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9323>
This reverts commit 6c8cc9be12.
A spec bug was resolved confirming the original behaviour. Also it
seems the game Foundation no longer depends on the incorrect
behaviour.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9486>
In a perf trace translate_quads_uint2uint_last2last_prdisable() was
showing up as a huge hot spot. Digging through the assembly on arm64
found that the compiler wasn't doing any read caching. Specifically,
the generated code looked roughly like this:
out[j+0] = in[i+0];
out[j+1] = in[i+1];
out[j+2] = in[i+3];
out[j+3] = in[i+1];
out[j+4] = in[i+2];
out[j+5] = in[i+3];
...and the compiler was loading "i+1" and "i+3" from memory twice for
no reason (instead of caching it).
If we sprinkle generous amounts of the `__restrict` keyword then the
compiler is able to be much smarter. Not only does it avoid
double-loading but it also generates better instructions. It uses two
LDRD instructions instead of 6 LDR instructions and uses some STRD
too.
In one example test this increased FPS from ~25.7 to ~34.5.
Change-Id: I88bf8bd9ac421fe48a7d6961e224425c3ae7beee
Reported-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9485>
We also switch from using __alignof__ to alignof() in util/macros.h
which works on MSVC with the one unfortunate downside of requiring an
actual type and not a value.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9511>