expand_asm_operands bug in 2.8.0

expand_asm_operands bug in 2.8.0

[Bernd Schmidt]

There was a change in GCC 2.8.0 to treat assembler statements with no
outputs as volatile. Unfortunately, it causes assembler statements with
exactly one output to be treated as volatile as well. In recent
Linux 2.1.x kernels, this causes rather bad code to be generated at times.

The patch below applies to GCC 2.8.0. For egcs, one part of it already exists
in the source, only the 2.8.0 method has to be deleted.

Bernd

	* stmt.c (expand_asm_operands): Undo change to treat assembler
	statements with no operands as volatile; use a different patch from
	egcs instead.

*** ./stmt.c.orig-1	Mon Feb  2 21:17:53 1998
--- ./stmt.c	Mon Feb  2 21:19:22 1998
*************** expand_asm_operands (string, outputs, in
*** 1421,1426 ****
--- 1421,1430 ----
    /* The insn we have emitted.  */
    rtx insn;
  
+   /* An ASM with no outputs needs to be treated as volatile.  */
+   if (noutputs == 0)
+     vol = 1;
+ 
    if (output_bytecode)
      {
        error ("`asm' is invalid when generating bytecode");
*************** expand_asm_operands (string, outputs, in
*** 1566,1574 ****
  		  TREE_STRING_POINTER (string), "", 0, argvec, constraints,
  		  filename, line);
  
!   /* The only use of BODY is if no outputs are specified, so set
!      it volatile, at least for now.  */
!   MEM_VOLATILE_P (body) = 1;
  
    /* Eval the inputs and put them into ARGVEC.
       Put their constraints into ASM_INPUTs and store in CONSTRAINTS.  */
--- 1570,1576 ----
  		  TREE_STRING_POINTER (string), "", 0, argvec, constraints,
  		  filename, line);
  
!   MEM_VOLATILE_P (body) = vol;
  
    /* Eval the inputs and put them into ARGVEC.
       Put their constraints into ASM_INPUTs and store in CONSTRAINTS.  */

Re: expand_asm_operands bug in 2.8.0

[Jeffrey A Law]

  In message < Pine.SOL.3.90.980203110130.16102H-100000@mccabe.informatik.rwth-aachen.de >you write:
  > There was a change in GCC 2.8.0 to treat assembler statements with no
  > outputs as volatile. Unfortunately, it causes assembler statements with
  > exactly one output to be treated as volatile as well. In recent
  > Linux 2.1.x kernels, this causes rather bad code to be generated at times.
  > 
  > The patch below applies to GCC 2.8.0. For egcs, one part of it already exis
  > ts
  > in the source, only the 2.8.0 method has to be deleted.
  > 
  > Bernd
  > 
  > 	* stmt.c (expand_asm_operands): Undo change to treat assembler
  > 	statements with no operands as volatile; use a different patch from
  > 	egcs instead.
Agreed and installed into egcs.

jeff