Latest addrlib supports SIMD (AVX2) and it's definitely fast enough to
be used in production now.
GFX10 is still not enabled by default due to some regressions from the
addrlib bump, also still missing AVX for some formats.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40996>
It's not yet enabled by default because it's quite a big extension
and I expect bugs because test coverage isn't very good. It will be
enabled by default in one or two Mesa releases when it's more stable.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39483>
The LLVM backend is unmaintained. Let's not encourage users to swap out
entire parts of the driver with an unsupported codepath. Enabling this
option is a footgun nowadays anyway, given that it disables many
features and thus may trigger bigger changes in behavior than intended.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40815>
Make it possible to trigger a emulated GPU reset by creating a file at
the path of the `LP_CONTEXT_RESET_FILE` env var. All llvmpipe contexts
using the `LOSE_CONTEXT_ON_RESET` strategy that where created before
the that file will start returning `PIPE_UNKNOWN_CONTEXT_RESET` in
`get_device_reset_status()`, while newer contexts - most notably those
created by apps in reaction to the device reset - continue to return
`PIPE_NO_RESET`.
Note that, because we're dealing with files, rtime is used instead of
mtime as the later
Usage:
```
LP_CONTEXT_RESET_FILE=~/llvmpipe_reset LIBGL_ALWAYS_SOFTWARE=1
someapp
touch ~/llvmpipe_reset
```
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40681>
Resizing the buffer isn't possible for per-submit captures and it
doesn't make sense either. This introduces a new envvar called
RADV_CACHE_COUNTERS_BUFFER_SIZE to increase the SPM BO size if needed.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40535>
Abusing RADV_PERFTEST for experimental features doesn't make real
sense, and I think we should stop doing that.
The existing RADV_PERFTEST options like RADV_PERFTEST=transfer_queue
still exists but they are marked as deprecated, they will be removed
in future Mesa releases.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40646>
The global BO list for app allocations has been enabled by default
since Mesa 25.3 and we didn't find any blockers, so let's make it the
default for real. Note that vkd3d-proton and Zink always used that
path and DXVK started to use it in August 2025 after requiring BDA.
This removes RADV_DEBUG=nobolist which was added only for debugging
purposes since the global BO list was enabled by default for app
allocations.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40466>
Add checks for integer coordinates and array indexing HW features
Features require HW support and the PCO_DEBUG env var to contain the
adv_smp entry
Integer coordinates are supported for images and textures without an LOD
setting
Array indexing is not supported and will trigger an abort
Reviewed-by: Simon Perretta <simon.perretta@imgtec.com>
Signed-off-by: Radu Costas <radu.costas@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40139>
Add support for the use of vertex input registers as additional general
purpose registers which previously was restricted to temporary
registers. Use of vertex input registers as additional general purpose
registers is not available for fragment shaders.
Vertex input registers are similar to temporary registers. The only
difference is that vertex input registers can contain pre-initialised
data when the shader starts.
By default, the number of vertex input registers used for register
allocation is the number of vertex input registers used for their
pre-initialised data rounded up to the nearest multiple of 4, as vertex
input registers are allocated in blocks of 4.
If PCO_DEBUG=alloc_extra_vtxins is used, a mimimum of 12 vertex input
registers are available for register allocation.
Signed-off-by: Duncan Brawley <duncan.brawley@imgtec.com>
Reviewed-by: Simon Perretta <simon.perretta@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39886>
Mesa now has a statistics framework. This adds support for emitting
additional statistics about PowerVR shaders for the Rogue architecture.
Add support for emitting the following statistics: Code size, scratch
size, spill count, temp count, loop count, number of inst groups, number
of main inst groups, number of bitwise inst groups and number of control
inst groups.
Add support for new PCO_DEBUG_PRINT option "stats" to emit shader stats.
Signed-off-by: Duncan Brawley <duncan.brawley@imgtec.com>
Reviewed-by: Simon Perretta <simon.perretta@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39523>
Similar to the low latency option for encode, this reduces latency
of decoding at the cost of increased power usage.
Can be enabled with AMD_DEBUG=lowlatencydec
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39450>
if set, creates a file in /tmp folder with mesa_<process_name>_<pid>_XXXXXX.log
logging all errors, warnings, etc., rather than stderr. The XXXXXX will be replaced
with alpha numeric character so for each run of the app a new log file will be
created guaranteed.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39206>
ACO got a lot better at forming VOPD instructions, and testing
feedback seems to point in a slightly positive direction for this.
gfx12 will also start requiring wave32 for dynamic VGPR allocation at
some point.
Measurements on navi31:
Cyberpunk 2077:
Difference at 95.0% confidence
1.12333 +/- 0.42876
1.88216% +/- 0.718391%
(Student's t, pooled s = 0.189165)
Black Myth Wukong benchmark:
Difference at 95.0% confidence
4 +/- 1.30862
13.9535% +/- 4.56495%
(Student's t, pooled s = 0.57735)
Portal with RTX:
66.2ms->61.5ms (~7.64% improvement)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39275>
Implement a mitigation for VM faults caused by SMEM reading
from NULL descriptors.
In order to satisfy VKD3D-Proton's expectations on mutable
descriptors, we must do this in shader code, it is not
sufficient to use the address of a mapped BO when writing
null descriptors. It is not feasible to mitigate this
in VKD3D-Proton.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38769>
RADV_PERFTEST=sparse is a new option to enable experimental
support for sparse features when they aren't enabled by default:
- gfx6 supports sparse, albeit with a reduced feature set
- gfx7 supports 3D images (with non-standard block shape)
and unaligned mip sizes
- gfx8 supports the same feature set as gfx7
(Polaris behaves more stable than other gfx8, so we had
already enabled it by default on Polaris for a long time.)
We pass all dEQP-VK.*sparse* tests on gfx6-8 when running on
a single thread however it may cause hangs or failures
when executing the tests on multiple parallel jobs.
We plan to enable this by default when we deem it stable enough.
Until then, users can already test some games that use it.
Note, at the moment there are some unsolved problems in the
amdgpu kernel driver regarding sparse bindings on these GPUs.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38553>
On Linux, drm-shim is the replacement.
On Windows, the project to support a compile-only device has been
abandonned since a while, so it's fine to not allow creating NULL
devices for now.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38544>
Add a new MESA_LOG_PREFIX environment variable that allows fine-grained
control over log prefixes in file logger output on Linux. The variable
accepts a comma-separated list of options:
- "tag": include the tag prefix (e.g., "MESA:")
- "level": include the level prefix (e.g., "info:")
By default, both tag and level are included. Users can customize the
prefix by setting MESA_LOG_PREFIX to any combination (e.g., "tag",
"level", "tag,level", or empty string for no prefix), making the output
more flexible and readable for different use cases.
Other loggers (syslog, Android logcat, Windows debugger) are unaffected
and continue to include tag and level information as appropriate for
their format.
Signed-off-by: Christian Gmeiner <cgmeiner@igalia.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38217>
Setting MESA_VK_VALIDATE_SHADER_BINARIES will cause the shader code to
round-trip every shader through [de]serialize and only ever use the
deserialized version. This catches bugs where the driver may drop
things in the [de]serialization process. It also deserializes the new
shader again and compares it against the original to ensure that
deserialize -> serialize is idempotent.
Acked-by: Eric R. Smith <eric.smith@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36647>
NIR+ACO is the best SSA-based shader compiler for AMD GPUs that exists.
There are many reasons why NIR+ACO is better than LLVM, and I have a long
list that I've collected over the years, but the major ones are better GPU
performance (faster GPU memory access thanks to better clauses and
scheduling, a lot less SGPR/VGPR spilling, better loop support, slightly
smaller shader binaries), 8x lower shader compile times, and smaller memory
footprint of the IR.
It also shows that NIR is a mature SSA-based shader compiler that helps
drivers generate optimized code very quickly.
And most importantly, radeonsi has slightly better Viewperf performance
with NIR+ACO than LLVM, and that's difficult to ignore.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38070>
Add documentation describing Android system property usage in Mesa. For
example, how environment varible names are translated by
os_get_option(), how to get/set values, and corresponding example
commands.
A new section is added to doc/envvars.rst which points to the full
details within the new "Android System Properties" section in
docs/android.rst.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37942>
vkd3d-proton (DX12) and Zink (GL) have always been enabling features
that require the global BO list to be enabled.
Since DXVK 2.7+ (August 2025), it's also always enabled by default for
DX9-11 games (because it requires BDA now).
The global BO list used to decrease performance in the past mostly
because of bad memory management in AMDGPU, but it seems the situation
slightly improved since. Though, there might still some workloads that
hit the issue, but I think it should be mostly good overall.
This introduces RADV_DEBUG=nobolist to disable the global BO list
when no features require it.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6957
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2331
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35075>
Instead of dumping multiple files with the optimizer passes, write a single
archive file with all the contents. The actual file is created
by the drivers, so later commits will actually enable the feature in
anv and iris.
This removes the use of INTEL_DEBUG=optimizer (and the corresponding
enum value) in brw. That environment variable is still used by ELK --
which currently doesn't support mda.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29146>
Some applications are not ready to handle multi plane
modifiers.
User who want this feature can use AMD_DEBUG=export_modifier
to enable it again.
Fixes: 0a266f0256 ("radeonsi: really support eglExportDMABUFImageQueryMESA")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13917
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37433>
The full HiZ workaround is the only one that fixes the issue reliably.
Sadly, the performance results are mixed and sometimes it hurts. To
maintain performance, RADV will opt-in by selecting which workarounds
to apply from drirc.
As we can't know the full list of games that will be affected by a
potential performance regression, users are encouraged to try with
RADV_GFX12_HIZ_WA (see Mesa documentation for more explanations).
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36174>
This lets us control how much logging we want Mesa to do, which is
useful in CI where we don't want to collect lots of log spam that then
has to be saved somewhere.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37203>
