target-sh4: improve shld instruction

The SH4 shld instruction can shift in both direction, depending on the
sign of the shift. This is currently implemented using branches, which
is not really efficient and prevents the optimizer to do its job. In
practice it is often used with a constant loaded in a register just
before.

Simplify the implementation by computing both the value shifted to the
left and to the right, and then selecting the correct one with a
movcond. As with a negative value the shift amount can go up to 32 which
is undefined, we shift the value in two steps.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>

1 files changed, 22 insertions, 26 deletions

diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index ca6ef5a..c8dd3a7 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -867,32 +867,28 @@ static void _decode_opc(DisasContext * ctx)
 	return;
     case 0x400d:		/* shld Rm,Rn */
 	{
-            TCGLabel *label1 = gen_new_label();
-            TCGLabel *label2 = gen_new_label();
-            TCGLabel *label3 = gen_new_label();
-	    TCGv shift;
-	    tcg_gen_brcondi_i32(TCG_COND_LT, REG(B7_4), 0, label1);
-	    /* Rm positive, shift to the left */
-            shift = tcg_temp_new();
-	    tcg_gen_andi_i32(shift, REG(B7_4), 0x1f);
-	    tcg_gen_shl_i32(REG(B11_8), REG(B11_8), shift);
-	    tcg_temp_free(shift);
-	    tcg_gen_br(label3);
-	    /* Rm negative, shift to the right */
-	    gen_set_label(label1);
-            shift = tcg_temp_new();
-	    tcg_gen_andi_i32(shift, REG(B7_4), 0x1f);
-	    tcg_gen_brcondi_i32(TCG_COND_EQ, shift, 0, label2);
-	    tcg_gen_not_i32(shift, REG(B7_4));
-	    tcg_gen_andi_i32(shift, shift, 0x1f);
-	    tcg_gen_addi_i32(shift, shift, 1);
-	    tcg_gen_shr_i32(REG(B11_8), REG(B11_8), shift);
-	    tcg_temp_free(shift);
-	    tcg_gen_br(label3);
-	    /* Rm = -32 */
-	    gen_set_label(label2);
-	    tcg_gen_movi_i32(REG(B11_8), 0);
-	    gen_set_label(label3);
+            TCGv t0 = tcg_temp_new();
+            TCGv t1 = tcg_temp_new();
+            TCGv t2 = tcg_temp_new();
+
+            tcg_gen_andi_i32(t0, REG(B7_4), 0x1f);
+
+            /* positive case: shift to the left */
+            tcg_gen_shl_i32(t1, REG(B11_8), t0);
+
+            /* negative case: shift to the right in two steps to
+               correctly handle the -32 case */
+            tcg_gen_xori_i32(t0, t0, 0x1f);
+            tcg_gen_shr_i32(t2, REG(B11_8), t0);
+            tcg_gen_shri_i32(t2, t2, 1);
+
+            /* select between the two cases */
+            tcg_gen_movi_i32(t0, 0);
+            tcg_gen_movcond_i32(TCG_COND_GE, REG(B11_8), REG(B7_4), t0, t1, t2);
+
+            tcg_temp_free(t0);
+            tcg_temp_free(t1);
+            tcg_temp_free(t2);
 	}
 	return;
     case 0x3008:		/* sub Rm,Rn */