[PULL,35/49] target/i386: optimize TEST+Jxx sequences

Commit Message

Paolo Bonzini Oct. 31, 2024, 5:51 p.m. UTC

Mostly used for TEST+JG and TEST+JLE, but it is easy to cover
also JBE/JA and JL/JGE; shaves about 0.5% TCG ops.

Reviewed-by: Richard Henderson <[e-mail]>
Signed-off-by: Paolo Bonzini <[e-mail]>
---
 target/i386/tcg/translate.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

Patch

diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c
index 5e326ab1aff..d3bbcf7317c 100644
--- a/target/i386/tcg/translate.c
+++ b/target/i386/tcg/translate.c
@@ -1069,6 +1069,28 @@  static CCPrepare gen_prepare_cc(DisasContext *s, int b, TCGv reg)
         }
         break;
 
+    case CC_OP_LOGICB ... CC_OP_LOGICQ:
+        /* Mostly used for test+jump */
+        size = s->cc_op - CC_OP_LOGICB;
+        switch (jcc_op) {
+        case JCC_BE:
+            /* CF = 0, becomes jz/je */
+            jcc_op = JCC_Z;
+            goto slow_jcc;
+        case JCC_L:
+            /* OF = 0, becomes js/jns */
+            jcc_op = JCC_S;
+            goto slow_jcc;
+        case JCC_LE:
+            /* SF or ZF, becomes signed <= 0 */
+            tcg_gen_ext_tl(cpu_cc_dst, cpu_cc_dst, size | MO_SIGN);
+            cc = (CCPrepare) { .cond = TCG_COND_LE, .reg = cpu_cc_dst };
+            break;
+        default:
+            goto slow_jcc;
+        }
+        break;
+
     default:
     slow_jcc:
         /* This actually generates good code for JC, JZ and JS.  */