Hello,

This patch avoids for sibling-tail-calls the use of pseudo-register.  Instead 
it uses for load of memory-address the accumulator-register.  By this we avoid 
that IRA/LRA need to choose a register.  So we reduce register-pressure.

The accumulator-register is always a valid register on tail-call case.  All 
other registers might be callee-saved, or used for argument-passing.  The only 
case where we would use the accumulator on call is the variadic-case for x86_64 
ABI.  Just that this function never is a candidate for sibling-tail-calls.

ChangeLog

2014-05-22  Kai Tietz  <kti...@redhat.com>

        * config/i386/i386.c (ix86_expand_call): Enforce for sibcalls
        on memory the use of accumulator-register.

Regression tested for x86_64-unknown-linux-gnu, x86_64-w64-mingw32, and 
i686-pc-cygwin.
Ok for apply?

Regards,
Kai

Index: i386.c
===================================================================
--- i386.c      (Revision 210412)
+++ i386.c      (Arbeitskopie)
@@ -24898,8 +24898,19 @@ ix86_expand_call (rtx retval, rtx fnaddr, rtx call
           ? !sibcall_insn_operand (XEXP (fnaddr, 0), word_mode)
           : !call_insn_operand (XEXP (fnaddr, 0), word_mode))
     {
+      rtx r;
       fnaddr = convert_to_mode (word_mode, XEXP (fnaddr, 0), 1);
-      fnaddr = gen_rtx_MEM (QImode, copy_to_mode_reg (word_mode, fnaddr));
+      if (!sibcall)
+       r = copy_to_mode_reg (word_mode, fnaddr);
+      else
+       {
+         r = gen_rtx_REG (word_mode, AX_REG);
+         if (! general_operand (fnaddr, VOIDmode))
+                   fnaddr = force_operand (fnaddr, r);
+         if (fnaddr != r)
+           emit_move_insn (r, fnaddr);
+       }
+      fnaddr = gen_rtx_MEM (QImode, r);
     }
 
   call = gen_rtx_CALL (VOIDmode, fnaddr, callarg1);

Reply via email to