diff mbox series

tcg/i386: Implement vector TST{EQ,NE} for avx512

Message ID 20240908185110.485573-1-richard.henderson@linaro.org
State New
Headers show
Series tcg/i386: Implement vector TST{EQ,NE} for avx512 | expand

Commit Message

Richard Henderson Sept. 8, 2024, 6:51 p.m. UTC
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---

Based-on: <20240908022632.459477-1-richard.henderson@linaro.org>
("tcg: Improve support for cmpsel_vec")

---
 tcg/i386/tcg-target.h     |  2 +-
 tcg/i386/tcg-target.c.inc | 31 ++++++++++++++++++++++++++++---
 2 files changed, 29 insertions(+), 4 deletions(-)

Comments

Philippe Mathieu-Daudé Sept. 9, 2024, 9:53 a.m. UTC | #1
On 8/9/24 20:51, Richard Henderson wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> 
> Based-on: <20240908022632.459477-1-richard.henderson@linaro.org>
> ("tcg: Improve support for cmpsel_vec")
> 
> ---
>   tcg/i386/tcg-target.h     |  2 +-
>   tcg/i386/tcg-target.c.inc | 31 ++++++++++++++++++++++++++++---
>   2 files changed, 29 insertions(+), 4 deletions(-)
> 
> diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
> index 342be30c4c..c68ac023d8 100644
> --- a/tcg/i386/tcg-target.h
> +++ b/tcg/i386/tcg-target.h
> @@ -224,7 +224,7 @@ typedef enum {
>   #define TCG_TARGET_HAS_minmax_vec       1
>   #define TCG_TARGET_HAS_bitsel_vec       have_avx512vl
>   #define TCG_TARGET_HAS_cmpsel_vec       1
> -#define TCG_TARGET_HAS_tst_vec          0
> +#define TCG_TARGET_HAS_tst_vec          have_avx512bw
>   
>   #define TCG_TARGET_deposit_i32_valid(ofs, len) \
>       (((ofs) == 0 && ((len) == 8 || (len) == 16)) || \
> diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
> index 8c363b7bfc..afeaab313a 100644
> --- a/tcg/i386/tcg-target.c.inc
> +++ b/tcg/i386/tcg-target.c.inc
> @@ -462,6 +462,14 @@ static bool tcg_target_const_match(int64_t val, int ct,
>   #define OPC_VPSRLVD     (0x45 | P_EXT38 | P_DATA16)
>   #define OPC_VPSRLVQ     (0x45 | P_EXT38 | P_DATA16 | P_VEXW)
>   #define OPC_VPTERNLOGQ  (0x25 | P_EXT3A | P_DATA16 | P_VEXW | P_EVEX)
> +#define OPC_VPTESTMB    (0x26 | P_EXT38 | P_DATA16 | P_EVEX)
> +#define OPC_VPTESTMW    (0x26 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
> +#define OPC_VPTESTMD    (0x27 | P_EXT38 | P_DATA16 | P_EVEX)
> +#define OPC_VPTESTMQ    (0x27 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
> +#define OPC_VPTESTNMB   (0x26 | P_EXT38 | P_SIMDF3 | P_EVEX)
> +#define OPC_VPTESTNMW   (0x26 | P_EXT38 | P_SIMDF3 | P_VEXW | P_EVEX)
> +#define OPC_VPTESTNMD   (0x27 | P_EXT38 | P_SIMDF3 | P_EVEX)
> +#define OPC_VPTESTNMQ   (0x27 | P_EXT38 | P_SIMDF3 | P_VEXW | P_EVEX)
>   #define OPC_VZEROUPPER  (0x77 | P_EXT)
>   #define OPC_XCHG_ax_r32	(0x90)
>   #define OPC_XCHG_EvGv   (0x87)
> @@ -3145,6 +3153,13 @@ static void tcg_out_cmp_vec_k1(TCGContext *s, TCGType type, unsigned vece,
>           { OPC_VPCMPB, OPC_VPCMPW, OPC_VPCMPD, OPC_VPCMPQ },
>           { OPC_VPCMPUB, OPC_VPCMPUW, OPC_VPCMPUD, OPC_VPCMPUQ }
>       };
> +    static const int testm_insn[4] = {
> +        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ
> +    };
> +    static const int testnm_insn[4] = {
> +        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ

OPC_VPTESTNMB, OPC_VPTESTNMW, OPC_VPTESTNMD, OPC_VPTESTNMQ ;)

Otherwise LGTM.

> +    };
> +
>       static const int cond_ext[16] = {
>           [TCG_COND_EQ] = 0,
>           [TCG_COND_NE] = 4,
> @@ -3160,9 +3175,19 @@ static void tcg_out_cmp_vec_k1(TCGContext *s, TCGType type, unsigned vece,
>           [TCG_COND_ALWAYS] = 7,
>       };
>   
> -    tcg_out_vex_modrm_type(s, cmpm_insn[is_unsigned_cond(cond)][vece],
> -                           /* k1 */ 1, v1, v2, type);
> -    tcg_out8(s, cond_ext[cond]);
> +    switch (cond) {
> +    case TCG_COND_TSTNE:
> +        tcg_out_vex_modrm_type(s, testm_insn[vece], /* k1 */ 1, v1, v2, type);
> +        break;
> +    case TCG_COND_TSTEQ:
> +        tcg_out_vex_modrm_type(s, testnm_insn[vece], /* k1 */ 1, v1, v2, type);
> +        break;
> +    default:
> +        tcg_out_vex_modrm_type(s, cmpm_insn[is_unsigned_cond(cond)][vece],
> +                               /* k1 */ 1, v1, v2, type);
> +        tcg_out8(s, cond_ext[cond]);
> +        break;
> +    }
>   }
>   
>   static void tcg_out_k1_to_vec(TCGContext *s, TCGType type,
Richard Henderson Sept. 9, 2024, 6:12 p.m. UTC | #2
On 9/9/24 02:53, Philippe Mathieu-Daudé wrote:
>> @@ -3145,6 +3153,13 @@ static void tcg_out_cmp_vec_k1(TCGContext *s, TCGType type, 
>> unsigned vece,
>>           { OPC_VPCMPB, OPC_VPCMPW, OPC_VPCMPD, OPC_VPCMPQ },
>>           { OPC_VPCMPUB, OPC_VPCMPUW, OPC_VPCMPUD, OPC_VPCMPUQ }
>>       };
>> +    static const int testm_insn[4] = {
>> +        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ
>> +    };
>> +    static const int testnm_insn[4] = {
>> +        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ
> 
> OPC_VPTESTNMB, OPC_VPTESTNMW, OPC_VPTESTNMD, OPC_VPTESTNMQ ;)

Oops, yes.  There are not so many uses of this in target/ yet,
and I happened to pick one that used testm.  :-/


r~
diff mbox series

Patch

diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index 342be30c4c..c68ac023d8 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -224,7 +224,7 @@  typedef enum {
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       have_avx512vl
 #define TCG_TARGET_HAS_cmpsel_vec       1
-#define TCG_TARGET_HAS_tst_vec          0
+#define TCG_TARGET_HAS_tst_vec          have_avx512bw
 
 #define TCG_TARGET_deposit_i32_valid(ofs, len) \
     (((ofs) == 0 && ((len) == 8 || (len) == 16)) || \
diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index 8c363b7bfc..afeaab313a 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -462,6 +462,14 @@  static bool tcg_target_const_match(int64_t val, int ct,
 #define OPC_VPSRLVD     (0x45 | P_EXT38 | P_DATA16)
 #define OPC_VPSRLVQ     (0x45 | P_EXT38 | P_DATA16 | P_VEXW)
 #define OPC_VPTERNLOGQ  (0x25 | P_EXT3A | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPTESTMB    (0x26 | P_EXT38 | P_DATA16 | P_EVEX)
+#define OPC_VPTESTMW    (0x26 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPTESTMD    (0x27 | P_EXT38 | P_DATA16 | P_EVEX)
+#define OPC_VPTESTMQ    (0x27 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPTESTNMB   (0x26 | P_EXT38 | P_SIMDF3 | P_EVEX)
+#define OPC_VPTESTNMW   (0x26 | P_EXT38 | P_SIMDF3 | P_VEXW | P_EVEX)
+#define OPC_VPTESTNMD   (0x27 | P_EXT38 | P_SIMDF3 | P_EVEX)
+#define OPC_VPTESTNMQ   (0x27 | P_EXT38 | P_SIMDF3 | P_VEXW | P_EVEX)
 #define OPC_VZEROUPPER  (0x77 | P_EXT)
 #define OPC_XCHG_ax_r32	(0x90)
 #define OPC_XCHG_EvGv   (0x87)
@@ -3145,6 +3153,13 @@  static void tcg_out_cmp_vec_k1(TCGContext *s, TCGType type, unsigned vece,
         { OPC_VPCMPB, OPC_VPCMPW, OPC_VPCMPD, OPC_VPCMPQ },
         { OPC_VPCMPUB, OPC_VPCMPUW, OPC_VPCMPUD, OPC_VPCMPUQ }
     };
+    static const int testm_insn[4] = {
+        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ
+    };
+    static const int testnm_insn[4] = {
+        OPC_VPTESTMB, OPC_VPTESTMW, OPC_VPTESTMD, OPC_VPTESTMQ
+    };
+
     static const int cond_ext[16] = {
         [TCG_COND_EQ] = 0,
         [TCG_COND_NE] = 4,
@@ -3160,9 +3175,19 @@  static void tcg_out_cmp_vec_k1(TCGContext *s, TCGType type, unsigned vece,
         [TCG_COND_ALWAYS] = 7,
     };
 
-    tcg_out_vex_modrm_type(s, cmpm_insn[is_unsigned_cond(cond)][vece],
-                           /* k1 */ 1, v1, v2, type);
-    tcg_out8(s, cond_ext[cond]);
+    switch (cond) {
+    case TCG_COND_TSTNE:
+        tcg_out_vex_modrm_type(s, testm_insn[vece], /* k1 */ 1, v1, v2, type);
+        break;
+    case TCG_COND_TSTEQ:
+        tcg_out_vex_modrm_type(s, testnm_insn[vece], /* k1 */ 1, v1, v2, type);
+        break;
+    default:
+        tcg_out_vex_modrm_type(s, cmpm_insn[is_unsigned_cond(cond)][vece],
+                               /* k1 */ 1, v1, v2, type);
+        tcg_out8(s, cond_ext[cond]);
+        break;
+    }
 }
 
 static void tcg_out_k1_to_vec(TCGContext *s, TCGType type,