With this patch, we compile separately
- general shaders (raygen, miss, callable)
- closest-hit shaders
- traversal shader (incl. all intersection / any-hit shaders)
Each shader uses the following scheme:
if (shader_pc == shader_va) {
<shader code>
}
next = select_next_shader(shader_va)
jump next
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22096>
Neither RENDER_SURFACE_STATE nor VDENC_SURFACE_CONTROL_BITS have a
Tiled Resource Mode field anymore. The RENDER_SURFACE_STATE field
was also overlapping with the L1 Cache Control settings field.
This also drops the assignment of that field in isl, because we were
just explicitly setting it to NONE (0) which is already the default
value genxml packing will give us. That saves us some ifdefs.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23449>
The FCV feature is documented to occur on regular render writes. Blitter
writes don't fall under that category. So, if the destination resource
of a blitter write had the FCV aux usage, we would have a good reason to
change the aux usage of the access to plain CCS_E.
We don't actually need to write code to handle this. iris doesn't use
any compression on blitter writes. Make that obvious with an assert.
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23220>
The FCV feature is documented to occur on regular render writes. Images
are written to with the DC data port however.
By using plain CCS_E for image writes, we can avoid the COMPRESSED_CLEAR
aux state in more cases. Doing this can avoid full resolves or partial
resolves for future accesses.
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23220>
We're going to start toggling between FCV_CCS_E and CCS_E. When
switching aux modes, flush_previous_aux_mode would typically perform
cache flushes for good reason. In the case of switching between CCS_E
with FCV on vs off, we haven't found aux mode flushing to matter. Treat
both CCS_E variants as equivalent to avoid extra cache flushing.
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23220>
Rename the isl_aux_usage enum to clarify that it is optional on gfx125.
The new name comes from the Alchemist docs, where the feature is
referred to as "Fast Clear Optimization (FCV)".
The rename was done with this command:
git grep -l "GFX12_CCS_E" | xargs sed -ie "s/GFX12_CCS_E/FCV_CCS_E/g"
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23220>
The 8 bpp restriction isn't specific to the automatic fast clear
behavior of GFX12_CCS_E. It is applicable to both GFX12_CCS_E and CCS_E.
Drop the aux usage check completely.
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23220>
Use the string itself as a key for searching -- and the internal
allocated name as a key when storing.
Because record_key_hash doesn't consider the name field, which is
the only used field for a SUBROUTINE type, the hash key was always
the same for all types. Using the name fixes this.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23277>
The original lineage for 14010455700 is 1808121037. Use this defect
number to apply the workaround to relevant platforms with the new
workaround mechanism.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23380>
The original lineage for 14010455700 is 1808121037. Use this defect
number to apply the workaround to relevant platforms with the new
workaround mechanism.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23380>
We can remove the LLVM 13 Wave32 discard workaround and
SI_PROFILE_IGNORE_LLVM13_DISCARD_BUG that disabled the workaround.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23471>
The demote emulation can be removed, and FS_CORRECT_DERIVS_AFTER_KILL
can be removed because it's always enabled on LLVM >= 13.
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23471>
rearrange the if checks so that there is less if checks
for newer gpus.
v2: fix block comment coding guideline (Marek Olšák)
simplify gfx9 if condition (Marek Olšák)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23454>
With num_cu = 4 in gfx11 measured power for idle, video playback and observed
power savings, hence enable dcc with retile for gfx11 with num_cu >= 4.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23454>
GPUs without native integers lower idiv in lower_int_to_float and
there is no need to call nir_lower_idiv(..) for such GPUs.
Fixes nir crashes I am seeing with gc2000_gles2 CI job.
Fixes: f532202f2d ("etnaviv: use nir_lower_idiv(..) before opt loop")
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23450>
If the viewport center is not positive we can't express it
through the fine coordinates, which are unsigned, and we
need to use the coarse coordinates too.
Fixes new crashes in Vulkan CTS 1.3.6.0:
dEQP-VK.draw.renderpass.offscreen_viewport.*negative*
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23489>
These are split between fine and coarse coordinates. We have only been
using fine until now, so we kept the same naming convention we had
prior to V3D 4.1 for simplicity, but we will start using the coarse
coordinates soon too.
Also, the signedness was reversed: coarse coordinates are signed and
fine coordinates are unsigned.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23489>
KHR_maintenance1 allows clients to specify a negative
height, but not a negative width, so assert on that and
simplify the computations for the horizontal dimension.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23489>
Previously the USC temps count was incorrectly being used for the
PDS temps count.
It was causing:
csbgen/pvr_packet_helpers.h:79:
__pvr_uint: Assertion `v <= max' failed.
This fixes the assert being hit on:
dEQP-VK.ubo.random.basic_arrays.2
dEQP-VK.ubo.random.all_shared_buffer.3
dEQP-VK.ubo.random.all_shared_buffer.48
dEQP-VK.ubo.random.all_out_of_order_offsets.3
This does not fully fix the tests though.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Matt Coster <matt.coster@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23356>
