Re: [PULL 02/38] tcg/optimize: Fix s_mask computation for shifts

On 8/21/26 09:51, Michael Tokarev wrote:
> On 8/18/26 20:01, Richard Henderson wrote:
>> Skip s_mask computation for logical right shift.
>>
>> Cc: [email protected]
>> Fixes: 93a967fbb57 ("tcg/optimize: Propagate sign info for shifting")
>> Reported-by: Jacob Young <[email protected]>
>> Signed-off-by: Richard Henderson <[email protected]>
>> ---
>>   tcg/optimize.c                     | 11 ++++++++++-
>>   tests/tcg/i386/test-i386-opt-shr.c | 21 +++++++++++++++++++++
>>   2 files changed, 31 insertions(+), 1 deletion(-)
>>   create mode 100644 tests/tcg/i386/test-i386-opt-shr.c
>>
>> diff --git a/tcg/optimize.c b/tcg/optimize.c
>> index ed2ff32ed2..d12babad88 100644
>> --- a/tcg/optimize.c
>> +++ b/tcg/optimize.c
>> @@ -2656,8 +2656,17 @@ static bool fold_shift(OptContext *ctx, TCGOp *op)
>>           z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh);
>>           o_mask = do_constant_folding(op->opc, ctx->type, o_mask, sh);
>> -        s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh);
>> +        if (op->opc == INDEX_op_shr) {
>> +            /*
>> +             * Logical right shift will force the sign bit zero.
>> +             * Don't bother computing s_mask and let fold_masks
>> +             * recompute from z_mask.
>> +             */
>> +            return fold_masks_zo(ctx, op, z_mask, o_mask);
>> +        }
>> +
>> +        s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh);
> 
> In 10.0.x, things are (were) a bit different.  Does this back-port look sane? --
> https://gitlab.com/mjt0k/qemu/-/commit/77ad6491326b4d0b2907f0ed9d504d6f490e245c
Perfect, thanks.

r~