aco: Unset 16 and 24-bit flags from operands in apply_extract.

Consider the following sequence in a shader:
b = p_extract a
c = v_mad_u32_u16 b, X, 0

The optimizer applies extract, resulting in:
c = v_mad_u32_u16 a, X, 0 (correct)

Then it mistakenly turns that into:
c = v_mul_u32_u24 a, X, 0 (incorrect)

In this case, the p_extract is applied to v_mad_u32_u16 by
apply_extract. After this, we can no longer be sure that
the operands are still 16 or 24-bit, so we have to remove
this flag.

No Fossil DB changes.

Fixes: 54292e99c7844500314bfd623469c65adef954c5
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/12558>

diff --git a/src/amd/compiler/aco_optimizer.cpp b/src/amd/compiler/aco_optimizer.cpp
index 7f36568..4a95ef8 100644 (file)
--- a/src/amd/compiler/aco_optimizer.cpp
+++ b/src/amd/compiler/aco_optimizer.cpp
@@ -871,6 +871,9 @@ apply_extract(opt_ctx& ctx, aco_ptr<Instruction>& instr, unsigned idx, ssa_info&
          instr->vop3().opsel |= 1 << idx;
    }
 
+   instr->operands[idx].set16bit(false);
+   instr->operands[idx].set24bit(false);
+
    ctx.info[tmp.id()].label &= ~label_insert;
    /* label_vopc seems to be the only one worth keeping at the moment */
    for (Definition& def : instr->definitions)