[PATCH v2 14/36] tcg: Use tcg_constant_{i32, vec} with tcg vec expanders

[Richard Henderson <richard.henderson@linaro.org>]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op-vec.c | 63 ++++++++++++++++++++++++++----------------------
 1 file changed, 34 insertions(+), 29 deletions(-)

diff --git a/tcg/tcg-op-vec.c b/tcg/tcg-op-vec.c
index f3927089a7..655b3ae32d 100644
--- a/tcg/tcg-op-vec.c
+++ b/tcg/tcg-op-vec.c
@@ -233,25 +233,17 @@ void tcg_gen_mov_vec(TCGv_vec r, TCGv_vec a)
     }
 }
 
-#define MO_REG  (TCG_TARGET_REG_BITS == 64 ? MO_64 : MO_32)
-
-static void do_dupi_vec(TCGv_vec r, unsigned vece, TCGArg a)
-{
-    TCGTemp *rt = tcgv_vec_temp(r);
-    vec_gen_2(INDEX_op_dupi_vec, rt->base_type, vece, temp_arg(rt), a);
-}
-
 TCGv_vec tcg_const_zeros_vec(TCGType type)
 {
     TCGv_vec ret = tcg_temp_new_vec(type);
-    do_dupi_vec(ret, MO_REG, 0);
+    tcg_gen_mov_vec(ret, tcg_constant_vec(type, MO_8, 0));
     return ret;
 }
 
 TCGv_vec tcg_const_ones_vec(TCGType type)
 {
     TCGv_vec ret = tcg_temp_new_vec(type);
-    do_dupi_vec(ret, MO_REG, -1);
+    tcg_gen_mov_vec(ret, tcg_constant_vec(type, MO_8, -1));
     return ret;
 }
 
@@ -267,37 +259,50 @@ TCGv_vec tcg_const_ones_vec_matching(TCGv_vec m)
     return tcg_const_ones_vec(t->base_type);
 }
 
-void tcg_gen_dup64i_vec(TCGv_vec r, uint64_t a)
+void tcg_gen_dupi_vec(unsigned vece, TCGv_vec dest, uint64_t val)
 {
-    if (TCG_TARGET_REG_BITS == 32 && a == deposit64(a, 32, 32, a)) {
-        do_dupi_vec(r, MO_32, a);
-    } else if (TCG_TARGET_REG_BITS == 64 || a == (uint64_t)(int32_t)a) {
-        do_dupi_vec(r, MO_64, a);
-    } else {
-        TCGv_i64 c = tcg_const_i64(a);
-        tcg_gen_dup_i64_vec(MO_64, r, c);
-        tcg_temp_free_i64(c);
+    TCGType type = tcgv_vec_temp(dest)->base_type;
+
+    /*
+     * For MO_64 constants that can't be represented in tcg_target_long,
+     * we must use INDEX_op_dup2_vec.
+     */
+    if (TCG_TARGET_REG_BITS == 32) {
+        val = dup_const(vece, val);
+        if (val != deposit64(val, 32, 32, val) &&
+            val != (uint64_t)(int32_t)val) {
+            uint32_t vl = extract64(val, 0, 32);
+            uint32_t vh = extract64(val, 32, 32);
+            TCGArg al = tcgv_i32_arg(tcg_constant_i32(vl));
+            TCGArg ah = tcgv_i32_arg(tcg_constant_i32(vh));
+            TCGArg di = tcgv_vec_arg(dest);
+
+            vec_gen_3(INDEX_op_dup2_vec, type, MO_64, di, al, ah);
+            return;
+        }
     }
+
+    tcg_gen_mov_vec(dest, tcg_constant_vec(type, vece, val));
 }
 
-void tcg_gen_dup32i_vec(TCGv_vec r, uint32_t a)
+void tcg_gen_dup64i_vec(TCGv_vec dest, uint64_t val)
 {
-    do_dupi_vec(r, MO_REG, dup_const(MO_32, a));
+    tcg_gen_dupi_vec(MO_64, dest, val);
 }
 
-void tcg_gen_dup16i_vec(TCGv_vec r, uint32_t a)
+void tcg_gen_dup32i_vec(TCGv_vec dest, uint32_t val)
 {
-    do_dupi_vec(r, MO_REG, dup_const(MO_16, a));
+    tcg_gen_dupi_vec(MO_32, dest, val);
 }
 
-void tcg_gen_dup8i_vec(TCGv_vec r, uint32_t a)
+void tcg_gen_dup16i_vec(TCGv_vec dest, uint32_t val)
 {
-    do_dupi_vec(r, MO_REG, dup_const(MO_8, a));
+    tcg_gen_dupi_vec(MO_16, dest, val);
 }
 
-void tcg_gen_dupi_vec(unsigned vece, TCGv_vec r, uint64_t a)
+void tcg_gen_dup8i_vec(TCGv_vec dest, uint32_t val)
 {
-    do_dupi_vec(r, MO_REG, dup_const(vece, a));
+    tcg_gen_dupi_vec(MO_8, dest, val);
 }
 
 void tcg_gen_dup_i64_vec(unsigned vece, TCGv_vec r, TCGv_i64 a)
@@ -502,8 +507,8 @@ void tcg_gen_abs_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
             if (tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0) {
                 tcg_gen_sari_vec(vece, t, a, (8 << vece) - 1);
             } else {
-                do_dupi_vec(t, MO_REG, 0);
-                tcg_gen_cmp_vec(TCG_COND_LT, vece, t, a, t);
+                tcg_gen_cmp_vec(TCG_COND_LT, vece, t, a,
+                                tcg_constant_vec(type, vece, 0));
             }
             tcg_gen_xor_vec(vece, r, a, t);
             tcg_gen_sub_vec(vece, r, r, t);
-- 
2.20.1