This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch: Installed preliminary branch pred/block reordering code.
- To: Jason Eckhardt <jle at cygnus dot com>, egcs-patches at egcs dot cygnus dot com
- Subject: Re: Patch: Installed preliminary branch pred/block reordering code.
- From: grahams <grahams at rcp dot co dot uk>
- Date: Fri, 14 Jan 2000 02:29:45 +0000
- References: <Pine.SUN.3.91.1000113180247.11015A-100000@rtl.cygnus.com>
Jason
Jason Eckhardt wrote:
> + /* EQ tests are usually false and NE tests are usually true. Also,
> + most quantities are positive, so we can make the appropriate guesses
> + about signed comparisons against zero. */
> + switch (GET_CODE (cond))
> + {
> + case CONST_INT:
> + /* Unconditional branch. */
> + prob = REG_BR_PROB_BASE / 2;
> + case EQ:
> + prob = REG_BR_PROB_BASE / 10;
> + case NE:
> + prob = REG_BR_PROB_BASE / 2;
> + case LE:
> + case LT:
> + if (XEXP (cond, 1) == const0_rtx)
> + prob = REG_BR_PROB_BASE / 10;
> + break;
> + case GE:
> + case GT:
> + if (XEXP (cond, 1) == const0_rtx
> + || (GET_CODE (XEXP (cond, 1)) == CONST_INT
> + && INTVAL (XEXP (cond, 1)) == -1))
> + prob = REG_BR_PROB_BASE / 2;
> + break;
> +
> + default:
> + prob = 0;
> + }
> + if (! find_reg_note (last_insn, REG_BR_PROB, 0))
> + REG_NOTES (last_insn)
> + = gen_rtx_EXPR_LIST (REG_BR_PROB, GEN_INT (prob),
> + REG_NOTES (last_insn));
> + }
> + }
> +
This above switch statement looks like its missing `break`s for
the `CONST_INT`,`EQ` and `NE' cases ?
Graham