This is the mail archive of the gcc@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]

Re: Your change to emit_group_{load,store}


On Oct 26, 2004, at 4:57 PM, Aldy Hernandez wrote:

Regarding Pinski's PR: 18163

The problem lies in that the old code was working for the following
parallel:

(parallel:TF [
        (expr_list:REG_DEP_TRUE (reg:DF 45 f13 [ g ])
            (const_int 0 [0x0]))

Now...can someone tell me what in the world that means?

No, it didn't get corrupted.  For that matter, the PPC back end created
it that way:

	  /* Describe where this piece goes in the fprs.  */
	  r = gen_rtx_REG (fmode, cum->fregno);
	  rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, const0_rtx);

Before I even venture a fix, I'd like to know what the above RTL is
supposed to mean.  Geoff?  David?

The notes establish a mapping between the registers of a multiple-word
parameter and the offsets in the parameter memory block that holds them.
(i.e. F13 gets stored at offset 0). The fact that it dumps as REG_DEP_TRUE
is a bug. Fariborz has fought with this quite a bit lately....

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