diff mbox series

[22/24] tcg/i386: Clear dest first in tcg_out_setcond if possible

Message ID 20230808031143.50925-23-richard.henderson@linaro.org
State Superseded
Headers show
Series tcg: Introduce negsetcond opcodes | expand

Commit Message

Richard Henderson Aug. 8, 2023, 3:11 a.m. UTC
Using XOR first is both smaller and more efficient,
though cannot be applied if it clobbers an input.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/i386/tcg-target.c.inc | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

Comments

Peter Maydell Aug. 11, 2023, 12:09 p.m. UTC | #1
On Tue, 8 Aug 2023 at 04:16, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Using XOR first is both smaller and more efficient,
> though cannot be applied if it clobbers an input.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM
diff mbox series

Patch

diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index e06ac638b0..cca49fe63a 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -1529,6 +1529,7 @@  static void tcg_out_setcond(TCGContext *s, int rexw, TCGCond cond,
                             int const_arg2)
 {
     bool inv = false;
+    bool cleared;
 
     switch (cond) {
     case TCG_COND_NE:
@@ -1578,9 +1579,23 @@  static void tcg_out_setcond(TCGContext *s, int rexw, TCGCond cond,
         break;
     }
 
+    /*
+     * If dest does not overlap the inputs, clearing it first is preferred.
+     * The XOR breaks any false dependency for the low-byte write to dest,
+     * and is also one byte smaller than MOVZBL.
+     */
+    cleared = false;
+    if (dest != arg1 && (const_arg2 || dest != arg2)) {
+        tgen_arithr(s, ARITH_XOR, dest, dest);
+        cleared = true;
+    }
+
     tcg_out_cmp(s, rexw, arg1, arg2, const_arg2, false);
     tcg_out_modrm(s, OPC_SETCC | tcg_cond_to_jcc[cond], 0, dest);
-    tcg_out_ext8u(s, dest, dest);
+
+    if (!cleared) {
+        tcg_out_ext8u(s, dest, dest);
+    }
 }
 
 #if TCG_TARGET_REG_BITS == 32