[PATCH 0/5] tcg: Support TCG_COND_TST* for vectors

[PATCH 0/5] tcg: Support TCG_COND_TST* for vectors

[Richard Henderson]

It would be weird if some TCG_COND_* codes are supported
in only restricted cases.  Let's go ahead and fill this out.

Based-on: 20240506010403.6204-1-richard.henderson@linaro.org
("[PATCH 00/57] target/arm: Convert a64 advsimd to decodetree (part 1)")

I will inject the target/arm patch as 45.5 in the above,
and route the rest through tcg-next.


r~


Richard Henderson (4):
  tcg: Introduce TCG_TARGET_HAS_tst_vec
  tcg: Expand TCG_COND_TST* if not TCG_TARGET_HAS_tst_vec
  tcg/aarch64: Support TCG_TARGET_HAS_tst_vec
  tcg/arm: Support TCG_TARGET_HAS_tst_vec
  target/arm: Use TCG_COND_TSTNE in gen_cmtst_vec

 include/tcg/tcg.h            |  1 +
 tcg/aarch64/tcg-target.h     |  1 +
 tcg/arm/tcg-target.h         |  1 +
 tcg/i386/tcg-target.h        |  1 +
 tcg/loongarch64/tcg-target.h |  1 +
 tcg/ppc/tcg-target.h         |  1 +
 tcg/s390x/tcg-target.h       |  1 +
 tcg/tcg-op-vec.c             | 18 ++++++++++++++++++
 tcg/aarch64/tcg-target.c.inc | 26 ++++++++++++++++++++++++--
 tcg/arm/tcg-target.c.inc     | 23 ++++++++++++++++++++---
 10 files changed, 69 insertions(+), 5 deletions(-)

-- 
2.34.1

[PATCH 1/5] tcg: Introduce TCG_TARGET_HAS_tst_vec

[Richard Henderson]

Prelude to supporting TCG_COND_TST* in vector comparisons.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/tcg/tcg.h            | 1 +
 tcg/aarch64/tcg-target.h     | 1 +
 tcg/arm/tcg-target.h         | 1 +
 tcg/i386/tcg-target.h        | 1 +
 tcg/loongarch64/tcg-target.h | 1 +
 tcg/ppc/tcg-target.h         | 1 +
 tcg/s390x/tcg-target.h       | 1 +
 7 files changed, 7 insertions(+)

diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h
index 2a1c080bab..21d5884741 100644
--- a/include/tcg/tcg.h
+++ b/include/tcg/tcg.h
@@ -155,6 +155,7 @@ typedef uint64_t TCGRegSet;
 #define TCG_TARGET_HAS_minmax_vec       0
 #define TCG_TARGET_HAS_bitsel_vec       0
 #define TCG_TARGET_HAS_cmpsel_vec       0
+#define TCG_TARGET_HAS_tst_vec          0
 #else
 #define TCG_TARGET_MAYBE_vec            1
 #endif
diff --git a/tcg/aarch64/tcg-target.h b/tcg/aarch64/tcg-target.h
index 85d5746e47..138bafb9da 100644
--- a/tcg/aarch64/tcg-target.h
+++ b/tcg/aarch64/tcg-target.h
@@ -167,6 +167,7 @@ typedef enum {
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       1
 #define TCG_TARGET_HAS_cmpsel_vec       0
+#define TCG_TARGET_HAS_tst_vec          0
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_NEED_LDST_LABELS
diff --git a/tcg/arm/tcg-target.h b/tcg/arm/tcg-target.h
index a43875cb09..434a892e07 100644
--- a/tcg/arm/tcg-target.h
+++ b/tcg/arm/tcg-target.h
@@ -150,6 +150,7 @@ extern bool use_neon_instructions;
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       1
 #define TCG_TARGET_HAS_cmpsel_vec       0
+#define TCG_TARGET_HAS_tst_vec          0
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_NEED_LDST_LABELS
diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index a10d4e1fce..2f67a97e05 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -224,6 +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_deposit_i32_valid(ofs, len) \
     (((ofs) == 0 && ((len) == 8 || (len) == 16)) || \
diff --git a/tcg/loongarch64/tcg-target.h b/tcg/loongarch64/tcg-target.h
index fede627bf7..29e4860d20 100644
--- a/tcg/loongarch64/tcg-target.h
+++ b/tcg/loongarch64/tcg-target.h
@@ -194,6 +194,7 @@ typedef enum {
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       1
 #define TCG_TARGET_HAS_cmpsel_vec       0
+#define TCG_TARGET_HAS_tst_vec          0
 
 #define TCG_TARGET_DEFAULT_MO (0)
 
diff --git a/tcg/ppc/tcg-target.h b/tcg/ppc/tcg-target.h
index 04a7aba4d3..e154fb14df 100644
--- a/tcg/ppc/tcg-target.h
+++ b/tcg/ppc/tcg-target.h
@@ -173,6 +173,7 @@ typedef enum {
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       have_vsx
 #define TCG_TARGET_HAS_cmpsel_vec       0
+#define TCG_TARGET_HAS_tst_vec          0
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_NEED_LDST_LABELS
diff --git a/tcg/s390x/tcg-target.h b/tcg/s390x/tcg-target.h
index ae448c3a3a..62ce9d792a 100644
--- a/tcg/s390x/tcg-target.h
+++ b/tcg/s390x/tcg-target.h
@@ -163,6 +163,7 @@ extern uint64_t s390_facilities[3];
 #define TCG_TARGET_HAS_minmax_vec     1
 #define TCG_TARGET_HAS_bitsel_vec     1
 #define TCG_TARGET_HAS_cmpsel_vec     0
+#define TCG_TARGET_HAS_tst_vec        0
 
 /* used for function call generation */
 #define TCG_TARGET_STACK_ALIGN		8
-- 
2.34.1

[PATCH 2/5] tcg: Expand TCG_COND_TST* if not TCG_TARGET_HAS_tst_vec

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op-vec.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/tcg/tcg-op-vec.c b/tcg/tcg-op-vec.c
index 094298bb27..84af210bc0 100644
--- a/tcg/tcg-op-vec.c
+++ b/tcg/tcg-op-vec.c
@@ -508,9 +508,11 @@ void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
     TCGTemp *rt = tcgv_vec_temp(r);
     TCGTemp *at = tcgv_vec_temp(a);
     TCGTemp *bt = tcgv_vec_temp(b);
+    TCGTemp *tt = NULL;
     TCGArg ri = temp_arg(rt);
     TCGArg ai = temp_arg(at);
     TCGArg bi = temp_arg(bt);
+    TCGArg ti;
     TCGType type = rt->base_type;
     int can;
 
@@ -518,6 +520,18 @@ void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
     tcg_debug_assert(bt->base_type >= type);
     tcg_assert_listed_vecop(INDEX_op_cmp_vec);
     can = tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece);
+
+    if (!TCG_TARGET_HAS_tst_vec && is_tst_cond(cond)) {
+        tt = tcg_temp_new_internal(type, TEMP_EBB);
+        ti = temp_arg(tt);
+        vec_gen_3(INDEX_op_and_vec, type, 0, ti, ai, bi);
+        at = tt;
+        ai = ti;
+        bt = tcg_constant_internal(type, 0);
+        bi = temp_arg(bt);
+        cond = tcg_tst_eqne_cond(cond);
+    }
+
     if (can > 0) {
         vec_gen_4(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
     } else {
@@ -526,6 +540,10 @@ void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
         tcg_expand_vec_op(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
         tcg_swap_vecop_list(hold_list);
     }
+
+    if (tt) {
+        tcg_temp_free_internal(tt);
+    }
 }
 
 static bool do_op3(unsigned vece, TCGv_vec r, TCGv_vec a,
-- 
2.34.1

[PATCH 3/5] tcg/aarch64: Support TCG_TARGET_HAS_tst_vec

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/aarch64/tcg-target.h     |  2 +-
 tcg/aarch64/tcg-target.c.inc | 26 ++++++++++++++++++++++++--
 2 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/tcg/aarch64/tcg-target.h b/tcg/aarch64/tcg-target.h
index 138bafb9da..8bd9e6a5eb 100644
--- a/tcg/aarch64/tcg-target.h
+++ b/tcg/aarch64/tcg-target.h
@@ -167,7 +167,7 @@ typedef enum {
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       1
 #define TCG_TARGET_HAS_cmpsel_vec       0
-#define TCG_TARGET_HAS_tst_vec          0
+#define TCG_TARGET_HAS_tst_vec          1
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_NEED_LDST_LABELS
diff --git a/tcg/aarch64/tcg-target.c.inc b/tcg/aarch64/tcg-target.c.inc
index 56fc9cb9e0..ffa8a3e519 100644
--- a/tcg/aarch64/tcg-target.c.inc
+++ b/tcg/aarch64/tcg-target.c.inc
@@ -2737,7 +2737,8 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
             TCGCond cond = args[3];
             AArch64Insn insn;
 
-            if (cond == TCG_COND_NE) {
+            switch (cond) {
+            case TCG_COND_NE:
                 if (const_args[2]) {
                     if (is_scalar) {
                         tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a1);
@@ -2752,7 +2753,27 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
                     }
                     tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
                 }
-            } else {
+                break;
+
+            case TCG_COND_TSTNE:
+            case TCG_COND_TSTEQ:
+                if (const_args[2]) {
+                    /* (x & 0) == 0 */
+                    tcg_out_dupi_vec(s, type, MO_8, a0,
+                                     -(cond == TCG_COND_TSTEQ));
+                    break;
+                }
+                if (is_scalar) {
+                    tcg_out_insn(s, 3611, CMTST, vece, a0, a1, a2);
+                } else {
+                    tcg_out_insn(s, 3616, CMTST, is_q, vece, a0, a1, a2);
+                }
+                if (cond == TCG_COND_TSTEQ) {
+                    tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a0);
+                }
+                break;
+
+            default:
                 if (const_args[2]) {
                     if (is_scalar) {
                         insn = cmp0_scalar_insn[cond];
@@ -2791,6 +2812,7 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
                     }
                     tcg_out_insn_3616(s, insn, is_q, vece, a0, a1, a2);
                 }
+                break;
             }
         }
         break;
-- 
2.34.1

[PATCH 4/5] tcg/arm: Support TCG_TARGET_HAS_tst_vec

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/arm/tcg-target.h     |  2 +-
 tcg/arm/tcg-target.c.inc | 23 ++++++++++++++++++++---
 2 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/tcg/arm/tcg-target.h b/tcg/arm/tcg-target.h
index 434a892e07..fb7261499b 100644
--- a/tcg/arm/tcg-target.h
+++ b/tcg/arm/tcg-target.h
@@ -150,7 +150,7 @@ extern bool use_neon_instructions;
 #define TCG_TARGET_HAS_minmax_vec       1
 #define TCG_TARGET_HAS_bitsel_vec       1
 #define TCG_TARGET_HAS_cmpsel_vec       0
-#define TCG_TARGET_HAS_tst_vec          0
+#define TCG_TARGET_HAS_tst_vec          1
 
 #define TCG_TARGET_DEFAULT_MO (0)
 #define TCG_TARGET_NEED_LDST_LABELS
diff --git a/tcg/arm/tcg-target.c.inc b/tcg/arm/tcg-target.c.inc
index 6a04c73c76..3de5f50b62 100644
--- a/tcg/arm/tcg-target.c.inc
+++ b/tcg/arm/tcg-target.c.inc
@@ -2740,17 +2740,33 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
     case INDEX_op_cmp_vec:
         {
             TCGCond cond = args[3];
+            ARMInsn insn;
 
-            if (cond == TCG_COND_NE) {
+            switch (cond) {
+            case TCG_COND_NE:
                 if (const_args[2]) {
                     tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a1);
                 } else {
                     tcg_out_vreg3(s, INSN_VCEQ, q, vece, a0, a1, a2);
                     tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
                 }
-            } else {
-                ARMInsn insn;
+                break;
 
+            case TCG_COND_TSTNE:
+            case TCG_COND_TSTEQ:
+                if (const_args[2]) {
+                    /* (x & 0) == 0 */
+                    tcg_out_dupi_vec(s, type, MO_8, a0,
+                                     -(cond == TCG_COND_TSTEQ));
+                    break;
+                }
+                tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a2);
+                if (cond == TCG_COND_TSTEQ) {
+                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
+                }
+                break;
+
+            default:
                 if (const_args[2]) {
                     insn = vec_cmp0_insn[cond];
                     if (insn) {
@@ -2769,6 +2785,7 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
                     tcg_debug_assert(insn != 0);
                 }
                 tcg_out_vreg3(s, insn, q, vece, a0, a1, a2);
+                break;
             }
         }
         return;
-- 
2.34.1

[PATCH 5/5] target/arm: Use TCG_COND_TSTNE in gen_cmtst_vec

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/tcg/gengvec.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/target/arm/tcg/gengvec.c b/target/arm/tcg/gengvec.c
index 1d6bc6021d..1895c3b19f 100644
--- a/target/arm/tcg/gengvec.c
+++ b/target/arm/tcg/gengvec.c
@@ -943,9 +943,7 @@ void gen_cmtst_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
 
 static void gen_cmtst_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
 {
-    tcg_gen_and_vec(vece, d, a, b);
-    tcg_gen_dupi_vec(vece, a, 0);
-    tcg_gen_cmp_vec(TCG_COND_NE, vece, d, d, a);
+    tcg_gen_cmp_vec(TCG_COND_TSTNE, vece, d, a, b);
 }
 
 void gen_gvec_cmtst(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
-- 
2.34.1

Re: [PATCH 1/5] tcg: Introduce TCG_TARGET_HAS_tst_vec

[Philippe Mathieu-Daudé]

On 15/5/24 16:58, Richard Henderson wrote:
> Prelude to supporting TCG_COND_TST* in vector comparisons.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   include/tcg/tcg.h            | 1 +
>   tcg/aarch64/tcg-target.h     | 1 +
>   tcg/arm/tcg-target.h         | 1 +
>   tcg/i386/tcg-target.h        | 1 +
>   tcg/loongarch64/tcg-target.h | 1 +
>   tcg/ppc/tcg-target.h         | 1 +
>   tcg/s390x/tcg-target.h       | 1 +
>   7 files changed, 7 insertions(+)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>