diff mbox

[RS6000] rs6000_preferred_reload_class

Message ID 20160804051434.GK20904@bubble.grove.modra.org
State New
Headers show

Commit Message

Alan Modra Aug. 4, 2016, 5:14 a.m. UTC 
On Wed, Aug 03, 2016 at 03:30:53PM -0400, Michael Meissner wrote:
> On Mon, Aug 01, 2016 at 09:00:43AM +0930, Alan Modra wrote:
> > Hi Mike,
> > I've been looking at a lot of reload/lra code lately in trying to fix
> > pr71680, and noticed a change to rs6000_preferred_reload_class.
> > 
> > In https://gcc.gnu.org/ml/gcc-patches/2016-05/msg00134.html you made
> > preferred_reload_class pick a register class when given NO_REGS as an
> > argument.  That seems odd to me given the usual behaviour of
> > preferred_reload_class is to restrict reg classes.  Did you mean to
> > make that change?  I'm wondering whether something you were playing
> > with escaped upstream, because there is no ChangeLog for it as far as
> > I can see..
> 
> I was playing with that, but given the patch was meant to be complex only, I
> suspect we should try taking it out now, and see if it affects anything.

Now bootstrapped and regression tested powerpc64le-linux and
powerpc64-linux.  OK to apply?

	* config/rs6000/rs6000.c (rs6000_preferred_reload_class): Delete
	code accidentally committed 2016-05-02 providing class when given
	NO_REGS.

Comments

Segher Boessenkool Aug. 4, 2016, 3:12 p.m. UTC | #1 
On Thu, Aug 04, 2016 at 02:44:34PM +0930, Alan Modra wrote:
> On Wed, Aug 03, 2016 at 03:30:53PM -0400, Michael Meissner wrote:
> > On Mon, Aug 01, 2016 at 09:00:43AM +0930, Alan Modra wrote:
> > > Hi Mike,
> > > I've been looking at a lot of reload/lra code lately in trying to fix
> > > pr71680, and noticed a change to rs6000_preferred_reload_class.
> > > 
> > > In https://gcc.gnu.org/ml/gcc-patches/2016-05/msg00134.html you made
> > > preferred_reload_class pick a register class when given NO_REGS as an
> > > argument.  That seems odd to me given the usual behaviour of
> > > preferred_reload_class is to restrict reg classes.  Did you mean to
> > > make that change?  I'm wondering whether something you were playing
> > > with escaped upstream, because there is no ChangeLog for it as far as
> > > I can see..
> > 
> > I was playing with that, but given the patch was meant to be complex only, I
> > suspect we should try taking it out now, and see if it affects anything.
> 
> Now bootstrapped and regression tested powerpc64le-linux and
> powerpc64-linux.  OK to apply?

Okay, thanks,


Segher


> 	* config/rs6000/rs6000.c (rs6000_preferred_reload_class): Delete
> 	code accidentally committed 2016-05-02 providing class when given
> 	NO_REGS.
diff mbox

Patch

diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 5a9ddd9..b923cfb 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -20441,25 +20441,6 @@  rs6000_preferred_reload_class (rtx x, enum reg_class rclass)
       return NO_REGS;
     }
 
-  /* If we haven't picked a register class, and the type is a vector or
-     floating point type, prefer to use the VSX, FPR, or Altivec register
-     classes.  */
-  if (rclass == NO_REGS)
-    {
-      if (TARGET_VSX && VECTOR_MEM_VSX_OR_P8_VECTOR_P (mode))
-	return VSX_REGS;
-
-      if (TARGET_ALTIVEC && VECTOR_MEM_ALTIVEC_P (mode))
-	return ALTIVEC_REGS;
-
-      if (DECIMAL_FLOAT_MODE_P (mode))
-	return TARGET_DFP ? FLOAT_REGS : NO_REGS;
-
-      if (TARGET_FPRS && TARGET_HARD_FLOAT && FLOAT_MODE_P (mode)
-	  && (reg_addr[mode].addr_mask[RELOAD_REG_FPR] & RELOAD_REG_VALID) != 0)
-	return FLOAT_REGS;
-    }
-
   if (GET_MODE_CLASS (mode) == MODE_INT && rclass == NON_SPECIAL_REGS)
     return GENERAL_REGS;