si_get_shader_variant_info doesn't need to check the kill flags because
killed stores are removed from NIR before that.
Only shader variants need to clear the writes_* flags if the epilog kills
them.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34492>
Just use nir_def_bits_used() instead of the manually written conditions.
These are gathered from bits of load_scalar_arg(vs_state_bits):
- uses_vs_state_indexed
- uses_gs_state_provoking_vtx_first
- uses_gs_state_outprim
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34492>
This is a step towards gathering shader info from shader variants instead of
input NIR.
uses_fbfetch_output can be ignored because it's already lowered to image
loads.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34492>
It shouldn't affect uses_vmem_load_other because divergent descriptors
are loaded with SMEM in waterfall loops.
Also all this removed code is highly questionable. Indirect access doesn't
matter for anything. Divergent access does, and that's handled correctly.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34492>
r600 is the only driver using this functionality, so let's move it there.
There's a few functions we used that aren't available outside of
u_blitter.c, but it turns out these aren't really needed. The first one
is blitter_check_saved_vertex_states(), that one just verifies that
certain states are saved as they should be. It's a bit sad to lose these
checks, but they are currently always passing, so nothing should break
by dropping them right now.
The second one is blitter_disable_render_cond(), and this one is also OK
to drop, because r600 never calls util_blitter_save_render_condition()
in the first place.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34954>
ls_sb_slot is unused now in cs_tracing_ctx and cs_exception_handler_ctx.
Let's remove it.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34808>
This commit moves cs_load_store_tracker from the cs_builder_conf
to cs_builder and makes it no longer optional to supply.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34808>
Progression logic is deprecated since v11 and we don't plan to use it.
Let's get ride of all increment logic on all instructions.
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Acked-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34947>
Evergreen and cayman have the hardware support for this extension.
This is described by the manual as EXPORT_RAT_INST_INC_UINT_RTN
and EXPORT_RAT_INST_DEC_UINT_RTN.
This change was tested and passes the piglit tests (17/17)
on cypress, palm and cayman.
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34501>
The assertion `assert(afence->seq_no <= *user_fence_cpu)` in
`amdgpu_fence_wait` can trigger a Mesa exit during GPU mode2 resets in
virtualized guest environments.
A GPU reset can cause the hardware to discard commands, including the
one that updates the user fence BO (`*user_fence_cpu`). This leaves
`*user_fence_cpu` with an older value, while `afence->seq_no` (from
command submission) is newer, leading to
`afence->seq_no > *user_fence_cpu` and triggering the assert.
Removing this assert prevents Mesa from exiting in this reset scenario.
No adverse side effects observed during testing. The assert appears
overly strict for hardware reset events where command completion is not
guaranteed.
Signed-off-by: Wei Zhao <wei.zhao@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34924>
It's only needed for old UVD without session context. Also remove
the NULL check for session context buffer as we always have the buffer.
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34807>
The maximal length is 65535 as an offset of 16 bits is being used to encode it.
Afterwards in VIRGL_CMD0, the buf_len equals 65536, so buf_len << 16 overflows its type which is uint32_t.
CID: 1604743 Overflowed constant
Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34816>
Do not override the handle number with 0 if we fail to create a new resource.
Also make sure to store the handle consistently in an uint32_t.
CID: 1644460 Overflowed constant
Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34816>
This handles basic operations where clang promotes integers to 32 bits
according to the C99 spec in OpenCL C source code.
This is its own opt_algerbraic pass, because we don't wanna fight with
nir_lower_bit_size.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34641>
StCurrBefore and StCurrAfter are sorted by POC.
For LtCurr there is no defined order, so if there are multiple long term
references in a picture, we need to parse the slice header.
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34698>
It doesn't serve any purpose when `CI_TRON_TIMEOUT__BOOT_CYCLE__RETRIES`
is not set to 1 or more.
The two exception are `zink-radv-vangogh-valve` and
`radv-polaris10-vkcts` which do set `retries` > 0.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34899>
Calculating lod does not always happen with the same vector size as
the vector size of the coords, which led to a crash when using the
new minLod feature. Use same repacking as with explicit lod.
Fixes: a8b104d9bd
Reviewed-by: Brian Paul <brian.paul@broadcom.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34906>
This had two uses, neither of them very good. On s390x we used it to
store the fact of our s390-ness, which we can replace with
DETECT_ARCH_S390. On x64 we used it to communicate AMD-Zen-ness to the
L3 topology detection code, which we can replace with a local parameter.
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34820>
This fixes the piglit arb_framebuffer_no_attachments-query test
(asserted in debug builds, crashed in release builds). Passes
with this change.
The SVGA_MAX_FRAMEBUFFER_DEFAULT_SAMPLES value is only used to
dimension the svga_rasterizer_state::altRastIds[] array.
Signed-off-by: Brian Paul <brian.paul@broadcom.com>
Reviewed-by: Roland Scheidegger <roland.scheidegger@broadcom.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34888>
