radv: More traversal loop micro-optimizations.
With a focus on integer ops this time.
Overall stats (all q2rtx)
Totals from 7 (0.01% of 134913) affected shaders:
CodeSize: 208476 -> 208292 (-0.09%); split: -0.11%, +0.02%
Instrs: 38384 -> 38248 (-0.35%)
Latency: 805725 -> 803142 (-0.32%)
InvThroughput: 165906 -> 165376 (-0.32%)
Copies: 4936 -> 4905 (-0.63%)
PreSGPRs: 393 -> 430 (+9.41%)
Reduces the number of VALU instructions executed per iteration by ~4%
Edited by Bas Nieuwenhuizen