This is the mail archive of the gcc-patches@gcc.gnu.org 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]

[PATCH] Fix ppro cost of load+operation


Hi,

This is now actually modelled in the PPro DFA, so it can
go away.  Bootstrapped and tested on i686, OK?

Gr.
Steven


	* i386.c (ix86_adjust_cost): Don't increase the cost for
	load+operation for PROCESSOR_PENTIUMPRO, it is already
	modelled in the DFA description

Index: i386.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
retrieving revision 1.669
diff -c -3 -p -r1.669 i386.c
*** i386.c      22 May 2004 18:18:46 -0000      1.669
--- i386.c      1 Jun 2004 23:05:32 -0000
*************** ix86_adjust_cost (rtx insn, rtx link, rt
*** 12261,12274 ****

      case PROCESSOR_PENTIUMPRO:
        memory = get_attr_memory (insn);
-       dep_memory = get_attr_memory (dep_insn);
-
-       /* Since we can't represent delayed latencies of load+operation,
-        increase the cost here for non-imov insns.  */
-       if (dep_insn_type != TYPE_IMOV
-           && dep_insn_type != TYPE_FMOV
-           && (dep_memory == MEMORY_LOAD || dep_memory == MEMORY_BOTH))
-       cost += 1;

        /* INT->FP conversion is expensive.  */
        if (get_attr_fp_int_src (dep_insn))
--- 12261,12266 ----


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