cmd/compile: implement more optimizations on loong64

[xen0n]

This issue is mainly for tracking the implementation progress of various low-hanging fruits regarding loong64 optimizations.

There are many missed optimization chances on loong64. A quick survey on SSA intrinsics uncovers:

• runtime.publicationBarrier
  • dmb st on arm64
  • dbar 0x1A on LA64 v1.10+ (gracefully downgrading to dbar 0 on LA64 v1.00) CL 577515
• runtime.Bswap{32,64}
  • revb.{2w,d} on LA64 v1.00: CL 483357
• runtime/internal/sys.Prefetch{,Streamed}
  • preld on LA64 v1.00
• runtime/internal/atomic.{And,Or}
  • am{and,or}.d on LA64 v1.00: CL 482756
• ❗ math.{Trunc,Ceil,Floor,RoundToEven} not possible with LA64 v1.00
  • LA64 v1.00 frint.[sd] is not orthogonal: no fixed rounding mode variants (unlike e.g. ftintr{m,p,z,ne}).
• ❓ math.Round
  • frint.[sd] on LA64 v1.00 -- have to check if the rounding mode behavior is tolerable
• math.Abs
  • fabs.[sd] on LA64 v1.00: CL 580280
• math.Copysign
  • fcopysign.[sd] on LA64 v1.00: CL 580280
• math.FMA
  • f{,n}m{add,sub}.[sd] on LA64 v1.00: CL 483355
• math/bits.TrailingZeros{64,32} (ssa.OpCtz{64,32})
  • ctz.[wd] on LA64 v1.00: CL 479498
• math/bits.Len{64,32,} (ssa.OpBitLen{64,32})
  • clz.[wd] on LA64 v1.00: CL 483356
  • significant performance regression across the board, needs investigation confirmed to be micro-architecture quirk, alleviated somewhat by various alignment tricks
• math/bits.Reverse{64,32,8} (ssa.OpBitRev{64,32,8})
  • bitrev.{d,w,4b} on LA64 v1.00: CL 483656

We may want to implement (and preferably benchmark) all of the above.

cc @golang/loong64