aco: skip &-4 before SMEM

aco: skip &-4 before SMEM

The hardware ignores the low 2 bits. I'm not sure if they are ignored
before or after the address is calculated, but this optimization should be
cautious enough.

fossil-db (Sienna Cichlid):
Totals from 259 (0.19% of 134572) affected shaders:
SpillSGPRs: 1381 -> 1382 (+0.07%)
SpillVGPRs: 1783 -> 1782 (-0.06%); split: -0.67%, +0.62%
CodeSize: 1598612 -> 1596084 (-0.16%); split: -0.30%, +0.14%
Scratch: 180224 -> 179200 (-0.57%); split: -1.14%, +0.57%
Instrs: 284885 -> 284268 (-0.22%); split: -0.34%, +0.12%
Latency: 6585634 -> 6603388 (+0.27%); split: -0.48%, +0.75%
InvThroughput: 2638983 -> 2648474 (+0.36%); split: -0.58%, +0.94%
VClause: 6797 -> 6820 (+0.34%); split: -0.15%, +0.49%
SClause: 6569 -> 6574 (+0.08%); split: -1.11%, +1.19%
Copies: 50561 -> 50586 (+0.05%); split: -0.61%, +0.66%
Branches: 10058 -> 10062 (+0.04%); split: -0.01%, +0.05%

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/13755>