[PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[Richard Henderson]

There is no such thing as vector extract.

Fixes: 932522a9ddc1 ("tcg/optimize: Fold and to extract during optimize")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3036
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 62a128bc9b..3638ab9fea 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1454,7 +1454,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
     a_mask = t1->z_mask & ~t2->o_mask;
 
     if (!fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, a_mask)) {
-        if (ti_is_const(t2)) {
+        if (op->opc == INDEX_op_and && ti_is_const(t2)) {
             /*
              * Canonicalize on extract, if valid.  This aids x86 with its
              * 2 operand MOVZBL and 2 operand AND, selecting the TCGOpcode
-- 
2.43.0

Re: [PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[Peter Maydell]

On Fri, 18 Jul 2025 at 18:46, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> There is no such thing as vector extract.
>
> Fixes: 932522a9ddc1 ("tcg/optimize: Fold and to extract during optimize")
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3036
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/optimize.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/tcg/optimize.c b/tcg/optimize.c
> index 62a128bc9b..3638ab9fea 100644
> --- a/tcg/optimize.c
> +++ b/tcg/optimize.c
> @@ -1454,7 +1454,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
>      a_mask = t1->z_mask & ~t2->o_mask;
>
>      if (!fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, a_mask)) {
> -        if (ti_is_const(t2)) {
> +        if (op->opc == INDEX_op_and && ti_is_const(t2)) {
>              /*
>               * Canonicalize on extract, if valid.  This aids x86 with its
>               * 2 operand MOVZBL and 2 operand AND, selecting the TCGOpcod


How does the fold_masks_zosa stuff work for vector operations here?
The masks are only 64 bits but the value we're working with is
wider than that, right?

thanks
-- PMM

Re: [PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[Richard Henderson]

On 7/18/25 12:11, Peter Maydell wrote:
> On Fri, 18 Jul 2025 at 18:46, Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> There is no such thing as vector extract.
>>
>> Fixes: 932522a9ddc1 ("tcg/optimize: Fold and to extract during optimize")
>> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3036
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>   tcg/optimize.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/tcg/optimize.c b/tcg/optimize.c
>> index 62a128bc9b..3638ab9fea 100644
>> --- a/tcg/optimize.c
>> +++ b/tcg/optimize.c
>> @@ -1454,7 +1454,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
>>       a_mask = t1->z_mask & ~t2->o_mask;
>>
>>       if (!fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, a_mask)) {
>> -        if (ti_is_const(t2)) {
>> +        if (op->opc == INDEX_op_and && ti_is_const(t2)) {
>>               /*
>>                * Canonicalize on extract, if valid.  This aids x86 with its
>>                * 2 operand MOVZBL and 2 operand AND, selecting the TCGOpcod
> 
> 
> How does the fold_masks_zosa stuff work for vector operations here?
> The masks are only 64 bits but the value we're working with is
> wider than that, right?

For vectors, the known one/zero bits stem from immediate operands.  All vector immediates 
are dup_const, that is, some replication of uint64_t or smaller element.  There is no way 
to represent (__vector uin8_t){ 1,2,3,4,5,6,7,8,9,a,b,c,d,e,f } with tcg at the moment.

Thus we can treat this sort of simple vector optimization as replications of uint64_t. 
Any other operation resets the masks to "unknown" state.


r~

Re: [PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[Pierrick Bouvier]

On 7/18/25 10:38 AM, Richard Henderson wrote:
> There is no such thing as vector extract.
> 
> Fixes: 932522a9ddc1 ("tcg/optimize: Fold and to extract during optimize")
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3036
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   tcg/optimize.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tcg/optimize.c b/tcg/optimize.c
> index 62a128bc9b..3638ab9fea 100644
> --- a/tcg/optimize.c
> +++ b/tcg/optimize.c
> @@ -1454,7 +1454,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
>       a_mask = t1->z_mask & ~t2->o_mask;
>   
>       if (!fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, a_mask)) {
> -        if (ti_is_const(t2)) {
> +        if (op->opc == INDEX_op_and && ti_is_const(t2)) {
>               /*
>                * Canonicalize on extract, if valid.  This aids x86 with its
>                * 2 operand MOVZBL and 2 operand AND, selecting the TCGOpcode

This solves the failures observed, thanks.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Tested-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>

Re: [PATCH for-10.1] tcg/optimize: Don't fold INDEX_op_and_vec to extract

[Peter Maydell]

On Fri, 18 Jul 2025 at 18:46, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> There is no such thing as vector extract.
>
> Fixes: 932522a9ddc1 ("tcg/optimize: Fold and to extract during optimize")
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3036
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/optimize.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM