Extended Description

Minimal test case:

static void attribute((noinline)) print_acc(__vector_quad *a) {
vector unsigned char a0[4];
__builtin_mma_disassemble_acc((void *)(a0), a);
print_values(a0[0]);
}
void test(__vector_quad *a, vector unsigned char ac) {
__builtin_mma_xvf32ger(a, ac, ac);
print_acc(a);
}

To reproduce:
clang -O3 -mcpu=pwr10 t.c -S