architecture: 0.968
performance: 0.958
graphic: 0.919
arm: 0.894
files: 0.862
device: 0.848
assembly: 0.842
ppc: 0.833
peripherals: 0.829
socket: 0.826
permissions: 0.825
PID: 0.799
debug: 0.786
user-level: 0.781
kernel: 0.778
vnc: 0.774
register: 0.701
network: 0.701
hypervisor: 0.695
semantic: 0.687
boot: 0.686
mistranslation: 0.642
risc-v: 0.639
virtual: 0.630
VMM: 0.618
x86: 0.529
TCG: 0.480
i386: 0.421
KVM: 0.346

qemu-aarch64: Incorrect result for ssra instruction when using vector lengths of 1024-bit or higher.
Description of problem:
```
#include <arm_sve.h>
#include <stdio.h>

#define SZ 32

int main(int argc, char* argv[]) {
  svbool_t pg = svptrue_b64();
  uint64_t VL = svcntd();

  fprintf(stderr, "One SVE vector can hold %li uint64_ts\n", VL);

  int64_t sr[SZ], sx[SZ], sy[SZ];
  uint64_t ur[SZ], ux[SZ], uy[SZ];

  for (uint64_t i = 0; i < SZ; ++i) {
    sx[i] = ux[i] = 0;
    sy[i] = uy[i] = 1024;
  }

  for (uint64_t i = 0; i < SZ; i+=VL) {
    fprintf(stderr, "Processing elements %li - %li\n", i, i + VL - 1);

    svint64_t SX = svld1(pg, sx + i);
    svint64_t SY = svld1(pg, sy + i);
    svint64_t SR = svsra(SX, SY, 4);
    svst1(pg, sr + i, SR);

    svuint64_t UX = svld1(pg, ux + i);
    svuint64_t UY = svld1(pg, uy + i);
    svuint64_t UR = svsra(UX, UY, 4);
    svst1(pg, ur + i, UR);
  }

  for (uint64_t i = 0; i < SZ; ++i) {
    fprintf(stderr, "sr[%li]=%li, ur[%li]\n", i, sr[i], ur[i]);
  }

  return 0;
}
```
Steps to reproduce:
1. Build the above C source using "gcc -march=armv9-a -O1 ssra.c", can also use clang.
2. Run with "qemu-aarch64 -cpu max,sve-default-vector-length=64 ./a.out" and you'll see the expected result of 64 (signed and unsigned)
3. Run with "qemu-aarch64 -cpu max,sve-default-vector-length=128 ./a.out" and you'll see the expected result of 64 for unsigned but the signed result is 0. This suggests the emulation of SVE2 ssra instruction is incorrect for this and bigger vector lengths.
Additional information: