We just need to pass the uses_discard flag to the epilog.
The hw skips the export anyway. This will hang if SPI registers declare
an output format or KILL_ENABLE is set because those cases require
an export with done=1.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16885>
Fixes -amdgpu-atomic-optimizations=true option.
From CommandLine.h:
/// Reset all command line options to a state that looks as if they have
/// never appeared on the command line. This is useful for being able to parse
/// a command line multiple times (especially useful for writing tests).
void ResetAllOptionOccurrences();
/// Reset the command line parser back to its initial state. This
/// removes
/// all options, categories, and subcommands and returns the parser to a state
/// where no options are supported.
void ResetCommandLineParser();
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Fixes: 7e2874dc93 ("ac: reset LLVM command line parser")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16850>
radeonsi will use it. This can be removed again after radeonsi support
radv_nir_lower_abi like lower pass.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16418>
GLSL says that indexing in a samplers array should be done with
a dynamically uniform value.
If the app doesn't obey this constraint, it'll get incorrect
results.
But there's one case where the app might be correct and yet
get incorrect result: if 2 consecutive draws are used, with no
state changes in between, the hardware might run them in the
same wave and the app will get incorrect result as-if it was
using non-dynamically uniform index.
To prevent this, this commit takes advantage of the divergence
analysis pass - if the index is marked as divergent, then it will
have the same effect as using ACCESS_NON_UNIFORM in SPIR-V.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16709>
In some scenario ac_init_llvm_target may be called twice,
but the LLVM library won't have been unloaded between
the 2 calls, leading to a LLVM warning being printed.
Example pseudo-code to trigger this for radeonsi:
gbm_create_device();
eglInitialize();
eglTerminate();
gbm_device_destroy();
gbm_create_device();
eglInitialize();
eglTerminate();
gbm_device_destroy();
To avoid the warning message from LLVM, clear the command line
parser state before calling LLVMParseCommandLineOptions.
This might fix https://gitlab.freedesktop.org/mesa/mesa/-/issues/5960
This is done only on LLVM 12+ because it seems to break some apps
on LLVM 11 (there has been some work post LLVM 11 release to refactor
CommandLine.cpp, see 42f588f39c5c and the following commits).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16587>
This method isn't part of the C API but we can still use it and
avoid getting error messages from the command line parser:
mesa: for the [...]: may only occur zero or one times
We could call it at the beginning of ac_init_llvm_target but
this may hide some real bugs so let drivers call it.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16587>
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Acked-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Martin Roukala <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16604>
LLVM uses an implicit argument for the ring offsets and this lowering
was just broken.
This fixes tessellation and geometry on all generations with LLVM.
Fixes: 896a55f47d ("radv: Lower ABI in NIR for tess/ESGS/NGG shader arguments.")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16530>
We can't use UINT16_ABGR for the alpha channel. Always use 32_ABGR.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16509>
This merges them into si_llvm_load_intrinsic and reuses load_tess_varyings.
RADV only implemented 1 callback.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16467>
because LLVM 15 doesn't insert s_waitcnt before s_barrier.
Now, control_barrier shouldn't add any waits because memory_barrier is
responsible for it, so remove the wait.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16304>
add_src was ignored and i32_0 was always used.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Fixes: 1e49018ced ("amd: Add extra source to the mbcnt_amd NIR intrinsic.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16427>
If dual source blending is enabled, use export targets 21 and 22.
Also we have to swap odd/even lanes between export target 21 and 22.
Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohanmarimuthu@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16328>
If a processor is unsupported, LLVM chooses "tahiti", which hangs
everything except gfx6. Check for support manually.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16112>
This depends on the offset computation fix from:
"ac/llvm: remove inst_offset parameter from ac_build_buffer_store_dword"
v2: The instruction type is changed to MUBUF, which requires us to clear
DATA_FORMAT with ADD_TID_ENABLE.
Reviewed-by: Mihai Preda <mhpreda@gmail.com> (v1)
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com> (v1)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15966>
