To simplify both llvm and aco backend and remove unnecessary
workaround code where prim count is known to be not zero.
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/22381>
Previously, the workaround only covered compile-time zero, but
this is insufficient and can cause GPU hangs in RadeonSI when
NGG culling is enabled.
Fix this by handling runtime zero in the workaround.
Cc: mesa-stable
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22370>
Formatted loads always support vec3, so this code didn't do anything.
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22117>
VMEM loads promoted from vec3 to vec4 didn't trim the vector, thus
returning vec4 on gfx6 and vec3 on later generations, which callers
don't expect.
SMEM loads were adding an extra component on gfx6, causing same issues.
Fixes: 82919e2d - amd: lower subdword UBO loads in NIR
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8693
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22117>
Mainly for nir_load_smem_buffer_amd which pass i32 for this parameter.
Fixes: 8030fbcf16 ("nir,ac/llvm: add nir_load_smem_buffer_amd")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22043>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Acked-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16805>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Acked-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16805>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19399>
LLVM is unable to select instructions for num_channels > 4, so we
workaround that by manually splitting larger buffer loads.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21358>
Change ac_build_buffer_load_common and ac_build_tbuffer_load so
the use structurized load when the vindex argument is not NULL.
Adjust callers to match the new behaviour.
This fixes the load_buffer_amd intrinsic with index source.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21358>
It's illegal and LLVM always knows which intrinsics don't read memory.
This started failing IR validation with LLVM 16.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20146>
This reverts commit 940734630d.
radeonsi and radv now support opaque pointers.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19035>
This way we can pass the type information to LLVM when needed.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19035>
And adapt the only user of this function.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19035>
The cast was superfluous: GEP2 return value is a pointer to
the given type.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19035>
LLVM always believes that this instruction's upper bound is the wave
size, regardless of ac_set_range_metadata and regardless of whether
the add source is used.
As a workaround, emit an extra add instruction.
Cc: mesa-stable
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17706>
Disabling opaque pointers in LLVM doesn't fix all the issues but
it makes pointers non-opaque by default (eg LLVMPointerType()
returns a typed pointer).
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17361>
nir lowering already call this with wave id check, no need to
check inside ac_build_sendmsg_gs_alloc_req again.
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/17130>
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>
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>