Re: [PATCH] tcg/optimize: Fix constant folding of setcond

[Richard Henderson]

On 12/6/24 03:58, wannacu wrote:
> The `z_mask` field of TCGTemp argument needs to be
> properly set for the upcoming `fold_setcond_zmask` call
>
> This patch resolves issues with running some x86_64
> applications (e.g., FontForge, Krita) on riscv64
>
> Signed-off-by: wannacu <wannacu2049@gmail.com>
> ---
>   tcg/optimize.c                   |  3 +++
>   tests/tcg/x86_64/Makefile.target |  1 +
>   tests/tcg/x86_64/setcond.c       | 28 ++++++++++++++++++++++++++++
>   3 files changed, 32 insertions(+)
>   create mode 100644 tests/tcg/x86_64/setcond.c
>
> diff --git a/tcg/optimize.c b/tcg/optimize.c
> index e9ef16b3c6..e580b8d8b1 100644
> --- a/tcg/optimize.c
> +++ b/tcg/optimize.c
> @@ -834,6 +834,9 @@ static int do_constant_folding_cond1(OptContext *ctx, TCGOp 
> *op, TCGArg dest,
>                                ? INDEX_op_and_i32 : INDEX_op_and_i64);
>           TCGOp *op2 = tcg_op_insert_before(ctx->tcg, op, and_opc, 3);
>           TCGArg tmp = arg_new_temp(ctx);
> +        /* Set z_mask for the follwing `fold_setcond_zmask` call. */
> +        arg_info(tmp)->z_mask = (ctx->type == TCG_TYPE_I32
> +                                      ? UINT32_MAX : UINT64_MAX);

I was curious as to why this helped, when arg_new_temp was supposed to be doing exactly 
this.  It turns out we were incorrectly reusing an old temp, not allocating a new one.


r~