Hi, Sergey!

thanks for the patch! LGTM

From: Mike Pall <mike>

(cherry picked from commit 43d0a19158ceabaa51b0462c1ebc97612b420a2e)

The previous commit accidentally removes the check that fusing
optimization isn't performing across the call to the `table.clear()`.
This commit fixes the behaviour by adding the corresponding check that
depends on the first bit of `check` masks in the `noconflict()` routine.

Sergey Kaplun:
* added the description for the problem

Part of tarantool/tarantool#10709
---
 src/lj_asm_x86.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/lj_asm_x86.h b/src/lj_asm_x86.h
index d77087d6..a96bc2e7 100644
--- a/src/lj_asm_x86.h
+++ b/src/lj_asm_x86.h
@@ -118,7 +118,7 @@ static int noconflict(ASMState *as, IRRef ref, IROp conflict, int check)
   while (--i > ref) {
     if (ir[i].o == conflict)
       return 0;  /* Conflict found. */
-    else if ((check & 1) && ir[i].o == IR_NEWREF)
+    else if ((check & 1) && (ir[i].o == IR_NEWREF || ir[i].o == IR_CALLS))
       return 0;
     else if ((check & 2) && (ir[i].op1 == ref || ir[i].op2 == ref))
       return 0;