On 11/24/20 9:28 AM, Richard Henderson wrote:
>> + tcg_gen_extract_i64(t0, cpu_gpr[a->rt], 0, 16);
>> + tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rd], t0, 0, 16);
>> + tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rd], t0, 16, 16);
>> + tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rd], t0, 32, 16);
>> + tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rd], t0, 48, 16);
>
> Actually, this would be better as
>
> tcg_gen_ext16u_i64(t0, cpu_gpr[rt]);
> tcg_gen_muli_i64(cpu_gpr[a->rd], t0, dup_const(1, MO_16));
Hmm, while that's fine for 64-bit hosts (and ideal for x86_64), it's not ideal
for the 32-bit hosts we have left.
This can also be done with
// replicate lower 16 bits, garbage in upper 32.
tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rt],
cpu_gpr[a->rt], 16, 48);
// replicate lower 32 bits
tcg_gen_deposit_i64(cpu_gpr[a->rd], cpu_gpr[a->rd],
cpu_gpr[a->rd], 32, 32);
r~
