- Jan 24, 2024
-
-
Vivante NPUs can contain systolic arrays that can be used to perform several kinds of tensor transformations. Use these to offload the CPU. Acked-by:
Christian Gmeiner <cgmeiner@igalia.com> Part-of: <mesa/mesa!25714>
-
Vivante NPUs contain MAC arrays that can be used to perform convolution operations on tensors. This commit implements the Gallium functions that accept convolutions and executes them on these MAC arrays. Acked-by:
-
The blob just emits the VIVS_GL_API_MODE and VIVS_PA_SYSTEM_MODE state changes in a single job before any other jobs. Reviewed-by:
-
Reviewed-by:
-
Update to rnndb commit 8d7ee714cfe2. Reviewed-by:
Christian Gmeiner <christian.gmeiner@gmail.com> Part-of: <mesa/mesa!25714>
-
Teflon is a Gallium frontend that TensorFlow Lite can load to delegate the execution of operations in a neural network model. See docs for more. Acked-by:
Dave Airlie <airlied@redhat.com> Part-of: <mesa/mesa!25714>
-
These define a stable API for implementing delegates that can be used in a Gallium frontend. Acked-by:
-
Some of the fields that were being left uninitialized were actually being used later. Reviewed-by:
-
Tomeu Vizoso authored
I'm having trouble merging stuff because of: ++ curl -L --retry 4 -f --retry-all-errors --retry-delay 60 'http://192.168.40.161:8888/cache/?uri=https://s3.freedesktop.org/artifacts/mesa/mesa/1085412/mesa-arm64-default-debugoptimized.tar.zst' % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0 curl: (7) Failed to connect to 192.168.40.161 port 8888 after 21 ms: Couldn't connect to server Warning: Problem (retrying all errors). Will retry in 60 seconds. 4 retries Warning: left. Part-of: <mesa/mesa!27231>
-
Juan A. Suárez authored
This reverts commit c16645c0. Bring the farm back again. Signed-off-by:
Juan A. Suarez Romero <jasuarez@igalia.com> Part-of: <mesa/mesa!27225>
-
- Jan 23, 2024
-
-
`test_full_yaml_log` don't need to submit job, because it would need to replicate/mock more stuff, like the first stage init, which is not necessary to reproduce issues from the raw YAML log file. Signed-off-by:
Guilherme Gallo <guilherme.gallo@collabora.com> Part-of: <mesa/mesa!26995>
-
Hard resets should not occur during the test phase. Therefore, let's detect them through specific log messages and raise an exception for a known issue if it occurs. Without this detection, the job will continue running on both Gitlab and LAVA until a timeout occurs. Real case: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/53546660 Signed-off-by:
-
Part-of: <mesa/mesa!27178>
-
Part-of: <mesa/mesa!27178>
-
Part-of: <mesa/mesa!27178>
-
Signed-off-by:
Rob Clark <robdclark@chromium.org> Part-of: <mesa/mesa!27219>
-
Changing force_vk_vendor=-1 for Palworld makes Palworld run. Not sure yet why. Game does not use XeSS currently. Closes: mesa/mesa#10474 Reviewed-by:
Lionel Landwerlin <lionel.g.landwerlin@intel.com> Part-of: <mesa/mesa!27218>
-
Part-of: <mesa/mesa!27221>
-
Part-of: <mesa/mesa!27221>
-
Part-of: <mesa/mesa!27221>
-
Otherwise, opt_if_simplification() can attempt to insert an inot after a jump. Fixes RADV compilation of a Cyberpunk 2077 pipeline with PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT. Signed-off-by:
Rhys Perry <pendingchaos02@gmail.com> Reviewed-by:
-
Fixes RADV compilation of a Doom Eternal pipeline with PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT, because nir_opt_non_uniform_access was skipped and later passes don't expect non-uniform access. Signed-off-by:
Faith Ekstrand <faith.ekstrand@collabora.com> Reviewed-by:
-
Fixes compilation of a Doom Eternal shader with PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT. ac_nir_lower_resinfo() was not happening because it is predicated on uses_resource_info_query and no later optimization updated it. Signed-off-by:
Friedrich Vock <friedrich.vock@gmx.de> Reviewed-by:
Samuel Pitoiset <samuel.pitoiset@gmail.com> Cc: mesa-stable Part-of: <mesa/mesa!27195>
-
Launch with : $ MESA_VK_TRACE=rmv MESA_VK_TRACE_TRIGGER=/tmp/trig ./my_app In another terminal, trigger a capture : $ touch /tmp/trig The application with create a snapshot and print out : RMV capture saved to '/tmp/my_app_2024.01.19_10.56.33.rmv' Then just open it with RMV : ./RadeonMemoryVisualizer /tmp/my_app_2024.01.19_10.56.33.rmv Signed-off-by:
José Roberto de Souza <jose.souza@intel.com> Part-of: <!26843>
-
Signed-off-by:
-
Signed-off-by:
-
Reviewed-by:
-
Prefix everything with anv_ so it doesn't clash with other drivers if loaded in the same process. Signed-off-by:
-
Signed-off-by:
-
Signed-off-by:
-
Part-of: <!27213>
-
We've refactored the common lib and panvk so they no longer depend on panfrost_{device,bo}.{c,h}. Let's move those files to the gallium driver dir so we're not tempted to use them again. Signed-off-by:Boris Brezillon <boris.brezillon@collabora.com> Reviewed-by:
Constantine Shablya <constantine.shablya@collabora.com> Reviewed-by:
Erik Faye-Lund <erik.faye-lund@collabora.com> Part-of: <mesa/mesa!26698>
-
Now that the rest of the code has been designed to not use panfrost_{bo,device}, we can transition the remaining wrappers to pan_kmod_{bo,vm,dev}. Signed-off-by: -
We wrap panfrost_device's BOs with a panvk_priv_bo object and store the result in panvk_device. This way we'll be able to transition to explicit BO allocation without changing BO users code. Signed-off-by:
-
We don't need the panfrost_bo object which contains both the BO and its CPU/GPU mappings. We store the GPU address at bind time. We also have a hack for index buffers which are currently walked by the CPU to extract the min/max index. Add a comment to make sure this field goes away when indirect draw is fixed. While at it, keep a ref to the buffer object so we don't end up with a invalid deref (UAF) if the vulkan user does something silly like freeing the VkDeviceMemory object while the VkBuffer is still active. Flag this with a TODO to make sure we don't forget about it. Signed-off-by:
-
We don't need the panfrost_bo object which contains both the BO and its CPU/GPU mappings. We store the GPU address at bind time and store it in the pimage object. While at it, keep a ref to the buffer object so we don't end up with a invalid deref (UAF) if the vulkan user does something silly like freeing the VkDeviceMemory object while the VkImage is still active. Flag this with a TODO to make sure we don't forget about it. Signed-off-by:
-
Signed-off-by:
-
Signed-off-by:
-
Signed-off-by:
-
Signed-off-by:
-
