1. Patchew
  2. QEMU
  3. View series

[PATCH] tcg/loongarch64: Fix cmp_vec with TCG_COND_NE

Richard Henderson posted 1 patch 1 month ago
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/20260623140609.645445-1-richard.henderson@linaro.org
Maintainers: WANG Xuerui <git@xen0n.name>, Richard Henderson <richard.henderson@linaro.org>
tcg/loongarch64/tcg-target.c.inc | 39 +++++++++++++++++++++++---------
1 file changed, 28 insertions(+), 11 deletions(-)
[PATCH] tcg/loongarch64: Fix cmp_vec with TCG_COND_NE
Posted by Richard Henderson 1 month ago 
For NE we need to invert EQ, not swap operands.

Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/3589
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/loongarch64/tcg-target.c.inc | 39 +++++++++++++++++++++++---------
 1 file changed, 28 insertions(+), 11 deletions(-)

diff --git a/tcg/loongarch64/tcg-target.c.inc b/tcg/loongarch64/tcg-target.c.inc
index c3350c90fc..182dcfd5eb 100644
--- a/tcg/loongarch64/tcg-target.c.inc
+++ b/tcg/loongarch64/tcg-target.c.inc
@@ -2371,19 +2371,36 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
                 default:
                     g_assert_not_reached();
                 }
-                break;
-            }
-
-            insn = cmp_vec_insn[cond][lasx][vece];
-            if (insn == 0) {
-                TCGArg t;
-                t = a1, a1 = a2, a2 = t;
-                cond = tcg_swap_cond(cond);
-                insn = cmp_vec_insn[cond][lasx][vece];
-                tcg_debug_assert(insn != 0);
+            } else {
+                switch (cond) {
+                case TCG_COND_EQ:
+                case TCG_COND_LE:
+                case TCG_COND_LEU:
+                case TCG_COND_LT:
+                case TCG_COND_LTU:
+                    insn = cmp_vec_insn[cond][lasx][vece];
+                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a1, a2));
+                    break;
+                case TCG_COND_GE:
+                case TCG_COND_GEU:
+                case TCG_COND_GT:
+                case TCG_COND_GTU:
+                    insn = cmp_vec_insn[tcg_swap_cond(cond)][lasx][vece];
+                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a2, a1));
+                    break;
+                case TCG_COND_NE:
+                    /* ne -> not(eq) */
+                    insn = cmp_vec_insn[TCG_COND_EQ][lasx][vece];
+                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a1, a2));
+                    insn = lasx ? OPC_XVNOR_V : OPC_VNOR_V;
+                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a0, a0));
+                    break;
+                default:
+                    g_assert_not_reached();
+                }
             }
         }
-        goto vdvjvk;
+        break;
     case INDEX_op_add_vec:
         tcg_out_addsub_vec(s, lasx, vece, a0, a1, a2, const_args[2], true);
         break;
-- 
2.43.0
Re: [PATCH] tcg/loongarch64: Fix cmp_vec with TCG_COND_NE
Posted by Michael Tokarev 2 weeks, 3 days ago 
On 23.06.2026 17:06, Richard Henderson wrote:
> For NE we need to invert EQ, not swap operands.
> 
> Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/3589
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

This feels like a qemu-stable material.
I'm picking it up for the stable series,
please let me know if I should not.

Thanks,

/mjt

> diff --git a/tcg/loongarch64/tcg-target.c.inc b/tcg/loongarch64/tcg-target.c.inc
> index c3350c90fc..182dcfd5eb 100644
> --- a/tcg/loongarch64/tcg-target.c.inc
> +++ b/tcg/loongarch64/tcg-target.c.inc
> @@ -2371,19 +2371,36 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
>                   default:
>                       g_assert_not_reached();
>                   }
> -                break;
> -            }
> -
> -            insn = cmp_vec_insn[cond][lasx][vece];
> -            if (insn == 0) {
> -                TCGArg t;
> -                t = a1, a1 = a2, a2 = t;
> -                cond = tcg_swap_cond(cond);
> -                insn = cmp_vec_insn[cond][lasx][vece];
> -                tcg_debug_assert(insn != 0);
> +            } else {
> +                switch (cond) {
> +                case TCG_COND_EQ:
> +                case TCG_COND_LE:
> +                case TCG_COND_LEU:
> +                case TCG_COND_LT:
> +                case TCG_COND_LTU:
> +                    insn = cmp_vec_insn[cond][lasx][vece];
> +                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a1, a2));
> +                    break;
> +                case TCG_COND_GE:
> +                case TCG_COND_GEU:
> +                case TCG_COND_GT:
> +                case TCG_COND_GTU:
> +                    insn = cmp_vec_insn[tcg_swap_cond(cond)][lasx][vece];
> +                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a2, a1));
> +                    break;
> +                case TCG_COND_NE:
> +                    /* ne -> not(eq) */
> +                    insn = cmp_vec_insn[TCG_COND_EQ][lasx][vece];
> +                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a1, a2));
> +                    insn = lasx ? OPC_XVNOR_V : OPC_VNOR_V;
> +                    tcg_out32(s, encode_vdvjvk_insn(insn, a0, a0, a0));
> +                    break;
> +                default:
> +                    g_assert_not_reached();
> +                }
>               }
>           }
> -        goto vdvjvk;
> +        break;
>       case INDEX_op_add_vec:
>           tcg_out_addsub_vec(s, lasx, vece, a0, a1, a2, const_args[2], true);
>           break;
Re: [PATCH] tcg/loongarch64: Fix cmp_vec with TCG_COND_NE
Posted by Michael Tokarev 2 weeks, 3 days ago 
On 09.07.2026 08:45, Michael Tokarev wrote:
> On 23.06.2026 17:06, Richard Henderson wrote:
>> For NE we need to invert EQ, not swap operands.
>>
>> Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/3589
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> 
> This feels like a qemu-stable material.
> I'm picking it up for the stable series,
> please let me know if I should not.

And for 10.0.x "LTS" series, I'm also picking up 2 extra commits:

a079836005 tcg/loongarch64: Fix vec_val computation in 
tcg_target_const_match
911f7328e9 tcg/loongarch64: Improve constraints for TCG_CT_CONST_VCMP

which looks like an improvement too :)

Again, please let me know if I should not.

Thanks,

/mjt
Re: [PATCH] tcg/loongarch64: Fix cmp_vec with TCG_COND_NE
Posted by Richard Henderson 2 weeks, 2 days ago 
On 7/8/26 22:54, Michael Tokarev wrote:
> On 09.07.2026 08:45, Michael Tokarev wrote:
>> On 23.06.2026 17:06, Richard Henderson wrote:
>>> For NE we need to invert EQ, not swap operands.
>>>
>>> Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/3589
>>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>>
>> This feels like a qemu-stable material.
>> I'm picking it up for the stable series,
>> please let me know if I should not.
> 
> And for 10.0.x "LTS" series, I'm also picking up 2 extra commits:
> 
> a079836005 tcg/loongarch64: Fix vec_val computation in tcg_target_const_match
> 911f7328e9 tcg/loongarch64: Improve constraints for TCG_CT_CONST_VCMP
> 
> which looks like an improvement too :)
> 
> Again, please let me know if I should not.

Thanks.

r~

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.