Re: [PATCH RFC 2/2] LoongArch: BPF: Eliminate zero/sign extensions for 32-bit operations

On 2026/7/26 下午5:23, Hengqi Chen wrote:
> On Sat, Jul 11, 2026 at 6:17 PM Tiezhu Yang <[email protected]> wrote:
>>
>> Eliminate frequent and redundant zero-extensions and sign-extensions
>> for 32-bit operations. This prevents potential functional bugs caused
>> by incorrect extension selections and reduces runtime overhead.

...

>> @@ -741,28 +725,32 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx, bool ext
>>          switch (code) {
>>          /* dst = src */
>>          case BPF_ALU | BPF_MOV | BPF_X:
>> +               if (imm == 1) {
>> +                       emit_insn(ctx, lu32id, reg, 0);
> 
> reg ?
> 
>> +                       break;

...

>>          /* dst = dst & imm */
>>          case BPF_ALU | BPF_AND | BPF_K:
>> +               if (is_unsigned_imm12(imm)) {
>> +                       emit_insn(ctx, andi, dst, dst, imm);
>> +               } else {
>> +                       move_imm(ctx, t1, imm);
>> +                       emit_insn(ctx, and, dst, dst, t1);
>> +               }
>> +               emit_insn(ctx, addiw, dst, dst, 0);
> 
> break; ?

Thanks for your review.
As mentioned in the cover letter, this hasn't been tested.
I just want to confirm that this is the correct direction.

The long-term goal is to eliminate frequent and redundant
zero-extensions and sign-extensions, but there are still
some issues to sort out. I will send the formal patch once
all problems are fixed.

For now, I will send out the stable and verified patches
first this week.

Thanks,
Tiezhu