The legacy::PassManager is only required to run backend optimizations
and for code generation. It should be deprecated when the new PM
can handle code generation on its own.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30506>
LLVM optimizes umul_hi with a constant to v_mul_hi_i32_i24_e32 which isn't
always what we need here. This causes miscalculations. To prevent LLVM to
apply this optimization, we insert a optimization barrier.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11761
Suggested-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30810>
This is only known to fix the assembly code when num_atomics > 6, which is
not currently used.
The VGPRs are reordered to simplify the clobber constraint.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30063>
With LLVM-19, Module.h header is not being pulled, which results in
compile errors e.g.
src/amd/llvm/ac_llvm_helper.cpp:102:10: error: no matching function for call to ‘unwrap(LLVMOpaqueModule*&)’
102 | unwrap(module)->setTargetTriple(TM->getTargetTriple().getTriple());
| ~~~~~~^~~~~~~~
In file included from /mnt/b/yoe/master/build/tmp/work/x86_64-linux/mesa-native/24.0.7/recipe-sysroot-native/usr/include/llvm/IR/Type.h:18,
from /mnt/b/yoe/master/build/tmp/work/x86_64-linux/mesa-native/24.0.7/recipe-sysroot-native/usr/include/llvm/IR/DerivedTypes.h:23,
from /mnt/b/yoe/master/build/tmp/work/x86_64-linux/mesa-native/24.0.7/recipe-sysroot-native/usr/include/llvm/IR/InstrTypes.h:26,
from /mnt/b/yoe/master/build/tmp/work/x86_64-linux/mesa-native/24.0.7/recipe-sysroot-native/usr/include/llvm/Analysis/TargetLibraryInfo.h:14,
from ../mesa-24.0.7/src/amd/llvm/ac_llvm_helper.cpp:8:
Its getting the definition from llvm/IR/Type.h instead of Module.h and caused
confusion to compiler
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11424
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29993>
The semantics of discard differ between GLSL and HLSL and
their various implementations. Subsequently, numerous application
bugs occurred and SPV_EXT_demote_to_helper_invocation was written
in order to clarify the behavior. In NIR, we now have 3 different
intrinsics for 2 things, and while demote and terminate have clear
semantics, discard still doesn't and can mean either of the two.
This patch entirely removes nir_intrinsic_discard and
nir_intrinsic_discard_if and replaces all occurences either with
nir_intrinsic_terminate{_if} or nir_intrinsic_demote{_if} in the
case that the NIR option 'discard_is_demote' is being set.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27617>
This makes the function easier to use when radeon_info is not available.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29243>
Usually this is lowered in NIR, but GFX12 needs to use an intrinsic.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29466>
They have been lowered in nir
v2: Keep the _amd versions
v3: Fix if's with removed ops
Signed-off-by: Mike Lothian <mike@fireburn.co.uk>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29280>
Effects:
- multi-component subdword handling removed because it's lowered
- 3-dword loads selected correctly instead of 4-dword loads
- the failure of dEQP-GLES3.functional.buffer.copy.subrange.large_to_small
due to LLVM exposed by a future commit is mysteriously fixed by this
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29053>
Use the same code as the pointer type. It works with all types and works
with any vector, but we need to handle i1 and v3i16 as special cases,
otherwise LLVM fails when it sees them. The previous code only extracted
the first component, which is not what we want.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28607>
We'll need to clean this up later, but for now it's better to
have it in common code than in RadeonSI.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28425>
