Miscompile with __arm_locally_streaming with -march=armv8-a+sme

[efriedma-quic]

Consider:

__arm_locally_streaming void f(int *p, int n) {
  #pragma clang loop vectorize_width(4, scalable)
  for (int i = 0; i < n; ++i) {
    p[i]++;
  }
}

Compile with -march=armv8-a+sme. The resulting assembly starts with:

        sub     sp, sp, #96
        rdsvl   x9, #1
        lsr     x9, x9, #3
        str     x9, [sp, #16]
        mov     x9, x0
        bl      __arm_get_current_vg
[...]

The bl corrupts x30.

Demo at https://godbolt.org/z/Ga9zaWEen

Workarounds:

• Use -march=armv8-a+sve+sme or similar
• Use -mno-omit-leaf-frame-pointer

CC @sdesmalen-arm @MacDue

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Eli Friedman (efriedma-quic)

Consider:

__arm_locally_streaming void f(int *p, int n) {
  #pragma clang loop vectorize_width(4, scalable)
  for (int i = 0; i < n; ++i) {
    p[i]++;
  }
}

Compile with -march=armv8-a+sme. The resulting assembly starts with:

        sub     sp, sp, #<!-- -->96
        rdsvl   x9, #<!-- -->1
        lsr     x9, x9, #<!-- -->3
        str     x9, [sp, #<!-- -->16]
        mov     x9, x0
        bl      __arm_get_current_vg
[...]

The bl corrupts x30.

Demo at https://godbolt.org/z/Ga9zaWEen

Workarounds:

• Use -march=armv8-a+sve+sme or similar
• Use -mno-omit-leaf-frame-pointer

CC @sdesmalen-arm @MacDue

[MacDue]

I'll take a look into this later today 👍