Use _tl operations for 32-bit operands on 32-bit targets, and only go
through trunc and extu ops for 64-bit targets.  While the trunc/ext
ops should be pretty much free after optimization, the optimizer also
does not like having the same temporary used in multiple EBBs.
Therefore it is nicer to not use tmpN* unless necessary.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 target/i386/tcg/emit.c.inc | 37 +++++++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 12 deletions(-)

diff --git a/target/i386/tcg/emit.c.inc b/target/i386/tcg/emit.c.inc
index 98c4c9569ef..f5e44117eab 100644
--- a/target/i386/tcg/emit.c.inc
+++ b/target/i386/tcg/emit.c.inc
@@ -1348,7 +1348,8 @@ static void gen_MULX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 
     /* low part of result in VEX.vvvv, high in MODRM */
     switch (ot) {
-    default:
+    case MO_32:
+#ifdef TARGET_X86_64
         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
         tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
         tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
@@ -1356,13 +1357,15 @@ static void gen_MULX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
         tcg_gen_extu_i32_tl(cpu_regs[s->vex_v], s->tmp2_i32);
         tcg_gen_extu_i32_tl(s->T0, s->tmp3_i32);
         break;
-#ifdef TARGET_X86_64
-    case MO_64:
-        tcg_gen_mulu2_i64(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
-        break;
-#endif
-    }
 
+    case MO_64:
+#endif
+        tcg_gen_mulu2_tl(cpu_regs[s->vex_v], s->T0, s->T0, s->T1);
+        break;
+
+    default:
+        g_assert_not_reached();
+    }
 }
 
 static void gen_PALIGNR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
@@ -1765,14 +1768,24 @@ static void gen_PSLLDQ_i(DisasContext *s, CPUX86State *env, X86DecodedInsn *deco
 static void gen_RORX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 {
     MemOp ot = decode->op[0].ot;
-    int b = decode->immediate;
+    int mask = ot == MO_64 ? 63 : 31;
+    int b = decode->immediate & mask;
 
-    if (ot == MO_64) {
-        tcg_gen_rotri_tl(s->T0, s->T0, b & 63);
-    } else {
+    switch (ot) {
+    case MO_32:
+#ifdef TARGET_X86_64
         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
-        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b & 31);
+        tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, b);
         tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
+        break;
+
+    case MO_64:
+#endif
+        tcg_gen_rotri_tl(s->T0, s->T0, b);
+        break;
+
+    default:
+        g_assert_not_reached();
     }
 }
 
-- 
2.43.0

On 12/23/23 05:15, Paolo Bonzini wrote:
> Use _tl operations for 32-bit operands on 32-bit targets, and only go
> through trunc and extu ops for 64-bit targets.  While the trunc/ext
> ops should be pretty much free after optimization, the optimizer also
> does not like having the same temporary used in multiple EBBs.
> Therefore it is nicer to not use tmpN* unless necessary.
> 
> Signed-off-by: Paolo Bonzini<pbonzini@redhat.com>
> ---
>   target/i386/tcg/emit.c.inc | 37 +++++++++++++++++++++++++------------
>   1 file changed, 25 insertions(+), 12 deletions(-)

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

r~