This looks like a regression from df30123794 ("radv: use
ac_compute_surface"). Before that, the opt4Space addrlib flag was set
to true unless the image has FMASK (ac_compute_surface will similarly
only set that flag for images without FMASK).
This saves multiple gigabytes of VRAM on one of our games, and brings
its VRAM utilisation on RADV in line with AMDGPU-PRO and NVIDIA.
Signed-off-by: Alex Smith <asmith@feralinteractive.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
"...and stay dead!"
Rafael deleted this file in c2b5a26dc2
(i965: Convert SF_STATE to genxml.) but Marek accidentally brought it
back in commit e7a091936f (mesa: replace
ctx->Polygon._FrontBit with a helper function) when resolving conflicts.
It's not actually even compiled, but it's still here trolling people
into thinking it still exists and needs patching.
Translate the NIR variables directly to LLVM instead of lowering to a
TGSI-style giant array of vec4's and then back to a variable. This
should fix indirect dereferences, make shared variables more tightly
packed, and make LLVM's alias analysis more precise. This should fix an
upcoming Feral title, which has a compute shader that was failing to
compile because the extra padding made us run out of LDS space.
v2: Combine the previous two patches into one, only use this for shared
variables for now until LLVM becomes smarter.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Tested-by: Alex Smith <asmith@feralinteractive.com>
Otherwise, if a client gave us a list of modifiers that contained a
modifier we understand but which is not supported on the hardware, we
might return that one and then fail to create the image.
Reviewed-by: Daniel Stone <daniels@collabora.com>
This commit splits the mapping in half. The modifier_infos table now
only contains the modifier and the since_gen field. The tiling bits
have been moved into a table in tiling_to_modifier as that's the only
place it was ever used. The modifier_is_supported function now takes a
devinfo and does the since_gen check.
Reviewed-by: Daniel Stone <daniels@collabora.com>
This assert was removed in b0cc55f298 but
got added back in 1a43d774b6, probably by
accident.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Now that we have an actual aux_usage field, we no longer need the
complex logic of is_lossless_compressed in order to figure out if a
miptree is CCS_E compressed. As a side-effect, there is not longer any
need to overload MSAA_LAYOUT_CMS for CCS_E and we can stop doing so.
Reviewed-by: Chad Versace <chadversary@chromium.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
HiZ, like MCS and CCS_E, can compress more than just clear colors so we
want it turned on whenever the miptree is being used as a depth
attachment. It's theoretically possible for someone to create a depth
texture, upload data with glTexSubImage2D, and texture from it without
ever binding it as a depth target. If this happens, we would end up
wasting a bit of space by allocating a HiZ surface we never use.
However, this is rather unlikely out side of test cases, so we're better
off just allocating it up-front.
Reviewed-by: Chad Versace <chadversary@chromium.org>
We need this split for the same reason that we need the split for CCS:
intel_miptree_supports_hiz is called *before* we choose the actual
tiling. Adding a tiling_supports_hiz helper lets choose_aux_usage
more accurately decide whether or not to enable hiz. In particular,
this prevents us from enabling HiZ on linear depth buffers.
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Fixes piglit test crash when context creation fails.
v2: As suggested by Brian, move the init to st_create_context_priv()
Reviewed-by: Brian Paul <brianp@vmware.com>
Enable the capability if the DRM supports it.
Hook up mechanism to send and receive fence FD from the DRM.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Connect fence_get_fd, fence_create_fd, and fence_server_sync.
Implement the required functions in vmw_fence module.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Connect fence_get_fd, fence_create_fd, and fence_server_sync.
Return PIPE_CAP_NATIVE_FENCE_FD capability based on what the
winsys reports
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
The new interfaces will be used to enable
EGL_ANDROID_native_fence_sync.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
The timeout parameter is required to implement
EGL_ANDROID_native_fence_sync.
v2
* Replaced default timeout from 0 to PIPE_TIMEOUT_INFINITE
* Add more documentation to the new timeout parameter
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
From the Vulkan 1.0.53 spec VU for vkCreateImageView:
"image must have been created with a usage value containing at least
one of VK_IMAGE_USAGE_SAMPLED_BIT, VK_IMAGE_USAGE_STORAGE_BIT,
VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT, or
VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT"
We were missing VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT from out list.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable@lists.freedesktop.org
We were early returning and never created the NULL surface state.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: James Legg <jlegg@feralinteractive.com>
Cc: mesa-stable@lists.freedesktop.org
If the queue is full, util_queue_add_job will wait while bo_fence_lock is
held.
It pb_slab wants to reuse a buffer, it will lock the pb_slab mutex and
try to check BO fence busyness, but it has to wait for bo_fence_lock to get
released. Both bo_fence_lock and pb_slab mutex are locked now.
When the CS thread unreferences and releases a suballocated buffer,
it will try to lock the pb_slab mutex and has to wait. The CS thread
can't finish its job in order to free a queue slot and unblock
util_queue_add_job ==> deadlock.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Consider the following situation:
mtx_lock(mutex);
do_something();
util_queue_add_job(...);
mtx_unlock(mutex);
If the queue is full, util_queue_add_job will wait for a free slot.
If the job which is currently being executed tries to lock the mutex,
it will be stuck forever, because util_queue_add_job is stuck.
The deadlock can be trivially resolved by increasing the queue size
(reallocating the queue) in util_queue_add_job if the queue is full.
Then util_queue_add_job becomes wait-free.
radeonsi will use it.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
During bring-up, this is often 0. Prevent automatic disablement of
ARB_timer_query and demotion of the OpenGL version to 3.2 by setting
a non-zero frequency. Print an error message instead.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
For inputs and outputs, indirect indexing is lowered by the GLSL compiler.
For temporaries, use alloca and disable the "promote-alloca" pass.
In the future, we could switch all codepaths to alloca permanently and
just rely on the "promote-alloca" pass.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Eric Engestrom
Samuel Pitoiset
Alex Smith
Dave Airlie
Kenneth Graunke
Connor Abbott
Jason Ekstrand
Charmaine Lee
Sinclair Yeh
Brian Paul
Marek Olšák