The first operand of v_bcnt should always be a VGPR because if it's
a SGPR, isel selects s_bcnt1 but I added a sanity check to prevent
any problems.
fossils-db (Vega10):
Totals from 23 (0.02% of 139517) affected shaders:
CodeSize: 106828 -> 106664 (-0.15%)
Instrs: 20242 -> 20201 (-0.20%)
Cycles: 213112 -> 211352 (-0.83%)
VMEM: 3200 -> 3184 (-0.50%)
SMEM: 928 -> 927 (-0.11%)
Helps Control, Assassins Creeds Origins and Youngblood.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7568>
Only on GFX8-9 because GFX10 doesn't zero the upper 16 bits.
No fossils-db changes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.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/7425>
No fossils-db changes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.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/7425>
v_mad_u32_u16 will be selected by isel to keep the range analysis
information around and to combine more v_add_u32+v_mad_u32_u16
together. When it's not possible to optimize that pattern, fallback
to v_mul_u32_u24 which is VOP2 instead of VOP3.
No fossils-db changes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.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/7425>
To indicate that the upper 16-bits are always 0 and that optimizing
v_mad_u32_u16 to v_mul_u32_u24 is valid.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.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/7425>
When one instruction doesn't fit into the existing labels, use
the generic one.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.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/7425>
Validation errors mention the pretty-printed instruction including
operands with the reserved % character, which caused vasprintf to
expect more format arguments than aco provided.
Fixes: c2b1978aa4 ("aco: rework the way various compilation/validation errors are reported")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7442>
The loop variable "k" shadowed another variable in the outer scope, so
this loop had no actual effect.
Fixes: 52cc1f8237 ("aco: improve p_create_vector RA for sub-dword operands")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7427>
I don't know why these values were introduced for but it seems like
we can optimize this by just doing:
gl_SampleMaskIn[0] = (SampleCoverage & (1 << gl_SampleID))
AMDGPU-PRO and AMDVLK apply the same formula to compute the
sample mask when per-sample shading is enabled.
No fossils-db changes.
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/7377>
Use auto to avoid mistyping the constness of the pair key, which
triggers implicit conversions rather than compilation errors.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7346>
Apply SGPRs/modifiers when possible and try not to break when SDWA
instructions are encountered.
No shader-db changes.
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/7349>
This seems to give no measurable benefit to Strange Brigade or Shadow of
Mordor, but it's simple to do, helps in theory and all other compilers do
it.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5919>
If the workgroup_size variable is lower than the actual workgroup size,
that means it's possible that ACO won't emit some s_barrier instructions
when in fact it should. This can possibly cause a GPU hang.
This is just for the sake of general correctness, currently this
can't cause a real problem because the maximum vertex count is always
greater than (or equal to) the primitive count in GS, and already
takes into account the number of GS invocations.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7232>
This is a workaround for a bug in Navi 1x NGG HW.
Very rarely, the Navi 1x PA can hang when an NGG workgroup exports
0 total primitives. According to AMD, we always need this workaround
when it is possible that the number of primitives is 0.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7232>
This is to make sure we don't compile a shader which doesn't
fit the available LDS space.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7232>
The p_reduce instruction only works if this operand is in a VGPR,
and otherwise gets lowered to incorrect code.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7232>
Example:
It is possible for some NGG GS waves to have 0 ES and/or GS invocations,
and in that case having an s_barrier inside divergent control flow can
very possibly hang the GPU.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7232>
Streams are really stateful and (IMO) difficult to read for non-trivial
usage. This is also more consistent with NIR and the rest of ACO.
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/7166>
This seems simpler to me. It should also work correctly when repeated
instructions cross blocks.
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/7166>
For non-arithmetic opcodes such as buffer_load_dword and buffer_load_short,
default to a definition size of 32.
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/7276>
This doesn't look like it would create correct IR for 8/16-bit phis and
doesn't seem to help anything. If we ever want to do this, it's probably
better done in nir_opt_remove_phis().
No fossil-db changes.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7216>
These are unused now that we almost always use p_parallelcopy for simple
copies.
No fossil-db changes.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7216>
