This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

RFA: fix pr21848 (i686-pc-linux-gnu bootstrap failure)

The documentation implies (although it doesn't actually spell it out) that we should have a valid
CALL_INSN_FUNCTION_USAGE field for ECF_CONST functions that read parameters on the stack.
This patch fixes emit_library_call_value_1 to actually adhere to this - we already have code in
expand_call to do this for user-defined ECF_CONST functions.

So far the only test that I have made is that the rtl and assembler looks plausible.
I'm bootstrapping this now on i686-pclinux-gnu, but I won't be able to see the results before tomorrow,
so if this is approved contingent on sucessful bootstrap / regtest, we can get i686-pc-linux-gnu faster
back to bootstrapple status if someone in another timezone tests this patch and checks it in.
2005-05-31  J"orn Rennecke <>

	* calls.c (emit_library_call_value_1): For const functions, add
	USEs of the stack slots to CALL_INSN_FUNCTION_USAGE.

Index: calls.c
RCS file: /cvs/gcc/gcc/gcc/calls.c,v
retrieving revision 1.389
diff -p -r1.389 calls.c
*** calls.c	9 May 2005 17:52:15 -0000	1.389
--- calls.c	31 May 2005 21:57:17 -0000
*************** emit_library_call_value_1 (int retval, r
*** 3629,3634 ****
--- 3629,3644 ----
  	      stack_usage_map[i] = 1;
+ 	  if (flags & ECF_CONST)
+ 	    {
+ 	      rtx use = plus_constant (argblock,
+ 				       argvec[argnum].locate.offset.constant);
+ 	      use = gen_rtx_MEM (argvec[argnum].mode, use);
+ 	      use = gen_rtx_USE (VOIDmode, use);
+ 	      call_fusage = gen_rtx_EXPR_LIST (VOIDmode, use, call_fusage);
+ 	    }

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]