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]

Re: cprop_reg problem on sparc

> Sorry, the final path component should be "mzero2.c"

Thanks.  I think we that need the same treatment in:

	  /* If we are accessing SRC in some mode other that what we
	     set it in, make sure that the replacement is valid.  */
	  if (mode != vd->e[regno].mode)
	      if (hard_regno_nregs[regno][mode]
		  > hard_regno_nregs[regno][vd->e[regno].mode])
		goto no_move_special_case;


  /* If we are narrowing the input to a smaller number of hard regs,
     and it is in big endian, we are really extracting a high part.
     Since we generally associate a low part of a value with the value itself,
     we must not do the same for the high part.
     Note we can still get low parts for the same mode combination through
     a two-step copy involving differently sized hard regs.
     Assume hard regs fr* are 32 bits bits each, while r* are 64 bits each:
     (set (reg:DI r0) (reg:DI fr0))
     (set (reg:SI fr2) (reg:SI r0))
     loads the low part of (reg:DI fr0) - i.e. fr1 - into fr2, while:
     (set (reg:SI fr2) (reg:SI fr0))
     loads the high part of (reg:DI fr0) into fr2.

     We can't properly represent the latter case in our tables, so don't
     record anything then.  */
  else if (sn < (unsigned int) hard_regno_nregs[sr][vd->e[sr].mode]
	   && (GET_MODE_SIZE (vd->e[sr].mode) > UNITS_PER_WORD

i.e. we need to bail out if we are narrowing and this is a big-endian target.

Eric Botcazou

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