Tags: xtaci/qpp

perf: optimize EncryptWithPRNG/DecryptWithPRNG for ~32% throughput im…

…provement

Performance improvement from ~1450 MB/s to ~1920 MB/s while maintaining
full backward compatibility.

Optimizations applied:
- Inline xoshiro256ss PRNG to eliminate function call overhead
- 2x loop unrolling (process 16 bytes per iteration instead of 8)
- Bounds check elimination hints using `_ = d[15]`
- Use bit shift `count << 3` instead of multiplication `count * 8`
- Pre-compute XOR offsets for better instruction-level parallelism
- Add //go:nosplit compiler directive to avoid stack split checks
- Early return for empty data slices
- Optimize struct field ordering for better cache locality
- Add padding to Rand struct for optimal alignment

Benchmark results (AMD Ryzen 9 5950X):
  Before: ~1450 MB/s
  After:  ~1920 MB/s

use modernized loop

encrypt full hmac instead of half hmac

use vet recommended writing style

Use (perms.BitLen()+7)/8 before deriving chunkCount.

fix fastprng

Rounded the chunk count up when deriving seed segments so seedToChunks

never under-fills the required entropy budget; also guard against
zero so there is always at least one 32-byte block (qpp.go:336-368).

add a fast PRNG for simple extension of seed, and move CreatePRNG to

public function

add a package description

optimize createprng