Re: [RFC PATCH v2 18/44] target/loongarch: Implement vsat

[Richard Henderson]

On 4/19/23 11:31, Song Gao wrote:
> 在 2023/4/1 下午1:03, Richard Henderson 写道:
> > Better to expand imm to max here, rather than both inside gen_vsat_s and the runtime 
> > do_vsats_*.
> >
> > Likewise for the unsigned versions. 
>
> I tried to expand imm to max  here  for the unsigned versions.
>
> {
>
>      uint64_t max;
>
>      ...
>
>      static const GVecGen2i op[4] = {
>          {
>              //.fniv = gen_vsat_u,
>              .fnoi = gen_helper_vsat_bu,
>              .opt_opc = vecop_list,
>              .vece = MO_8
>          },
>          {
>              //.fniv = gen_vsat_u,
>              .fnoi = gen_helper_vsat_hu,
>              .opt_opc = vecop_list,
>              .vece = MO_16
>          },
>          {
>              //.fniv = gen_vsat_u,
>              .fnoi = gen_helper_vsat_wu,
>              .opt_opc = vecop_list,
>              .vece = MO_32
>          },
>          {
>              //.fniv = gen_vsat_u,
>              .fnoi = gen_helper_vsat_du,
>              .opt_opc = vecop_list,
>              .vece = MO_64
>          },
>      };
>
>      max = (imm == 0x3f) ? UINT64_MAX : (1ull << (imm + 1)) - 1;
>      tcg_gen_gvec_2i(vd_ofs, vj_ofs, oprsz, maxsz, max, &op[vece]);
>
> }
>
>
> and  I got a tcg_debug_assert();
>
>
> Thread 1 "qemu-loongarch6" received signal SIGABRT, Aborted.
> 0x00007ffff60b337f in raise () from /lib64/libc.so.6
> (gdb) bt
> #0  0x00007ffff60b337f in raise () from /lib64/libc.so.6
> #1  0x00007ffff609ddb5 in abort () from /lib64/libc.so.6
> #2  0x00007ffff609dc89 in __assert_fail_base.cold.0 () from /lib64/libc.so.6
> #3  0x00007ffff60aba76 in __assert_fail () from /lib64/libc.so.6
> #4  0x0000555555632fcf in simd_desc (oprsz=16, maxsz=16, data=134217727) at 
> ../tcg/tcg-op-gvec.c:91

You should use tcg_gen_gvec_2s, and pass tcg_constant_i64(max).


r~