This is the mail archive of the
mailing list for the GCC project.
Re: [RS6000] Fix PR61098, Poor code setting count register
- From: Alan Modra <amodra at gmail dot com>
- To: David Edelsohn <dje dot gcc at gmail dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Fri, 9 May 2014 12:10:54 +0930
- Subject: Re: [RS6000] Fix PR61098, Poor code setting count register
- Authentication-results: sourceware.org; auth=none
- References: <20140508014846 dot GA5162 at bubble dot grove dot modra dot org> <CAGWvnyme6a+QLSxXwzUOxWAHWTokxZCySua_+25hUzaEVYvgPA at mail dot gmail dot com>
On Thu, May 08, 2014 at 09:48:35AM -0400, David Edelsohn wrote:
> The history is 32 bit HWI.
> The ChangeLog does not mention the changes to rs6000.md nor rs6000-protos.h.
* config/rs6000/rs6000.md (movsi_internal1_single+1): Update
call to rs6000_emit_set_const in splitter.
(movdi_internal64+2, +3): Likewise.
* config/rs6000/rs6000-protos.h (rs6000_emit_set_const): Update
> Please do not remove all of the comments from the two functions. The
> comments should provide some documentation about the different
> purposes of the two functions other than setting DEST to a CONST.
I believe my updated comment covers the complete purpose of the
function nowadays. The comments I removed are out-dated, and should
have been removed a long time ago.. rs6000_emit_set_const does not
even look at N, it always returns a non-zero result, and the return is
only tested for non-zero. I removed MODE too, because that is always
the same as GET_MODE (dest).
> Why did you remove the test for NULL dest?
> - if (dest == NULL)
> - dest = gen_reg_rtx (mode);
> That could occur, at least it used to occur.
I'm sure we can't get a NULL dest nowadays. All (three) uses of
rs6000_emit_set_const occur in splitters. They all must have passed a
gpc_reg_operand constraint on operands before calling
rs6000_emit_set_const, so if NULL were possible we'd segfault in
> I think that the way you rearranged the invocations of copy_rtx() in
> rs6000_emit_set_long_const() is okay, but it would be good for someone
> else to double check.
Yeah, that function is a bit messy. I took the approach of always use
a bare "dest" once in the last instruction emitted, with every other
use getting hit with copy_rtx. The previous approach was similar,
but used the bare "dest" on the first instruction emitted. Obviously
you don't need copy_rtx anywhere with the new code when
can_create_pseudo_p is true, but I felt it wasn't worth optimising
that for the added source complication.
Australia Development Lab, IBM