diff mbox

[rs6000] Prefer vnand/vperm over vnor/vperm for Power8

Message ID 1396984045.5401.125.camel@gnopaine
State New
Headers show

Commit Message

Bill Schmidt April 8, 2014, 7:07 p.m. UTC 
Hi,

Recently I posted a patch to use vnor to invert the permute control
vector for little-endian permutes.  It's come to my attention that, when
available, the vnand instruction should be used in preference to vnor
for possible future processor exploitation.  This patch checks for
availability of vnand and uses it instead.  It also removes earlier use
of emit_move_insn in this logic where that was not appropriate.

Bootstrapped and tested on powerpc64le-unknown-linux-gnu with no
regressions.  Is this ok for trunk?

I would also like to backport the combined patch to the FSF 4.8 branch,
if you feel that is appropriate.  Please let me know.

Thanks,
Bill


2014-04-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

	* config/rs6000/rs6000.c (rs6000_expand_vector_set): Use vnand
	instead of vnor to exploit possible fusion opportunity in the
	future.
	(altivec_expand_vec_perm_const_le): Likewise.

Comments

David Edelsohn April 8, 2014, 7:52 p.m. UTC | #1 
On Tue, Apr 8, 2014 at 3:07 PM, Bill Schmidt
<wschmidt@linux.vnet.ibm.com> wrote:
> Hi,
>
> Recently I posted a patch to use vnor to invert the permute control
> vector for little-endian permutes.  It's come to my attention that, when
> available, the vnand instruction should be used in preference to vnor
> for possible future processor exploitation.  This patch checks for
> availability of vnand and uses it instead.  It also removes earlier use
> of emit_move_insn in this logic where that was not appropriate.
>
> Bootstrapped and tested on powerpc64le-unknown-linux-gnu with no
> regressions.  Is this ok for trunk?
>
> I would also like to backport the combined patch to the FSF 4.8 branch,
> if you feel that is appropriate.  Please let me know.
>
> Thanks,
> Bill
>
>
> 2014-04-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
>
>         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Use vnand
>         instead of vnor to exploit possible fusion opportunity in the
>         future.
>         (altivec_expand_vec_perm_const_le): Likewise.

Okay for trunk and 4.8.

Thanks, David
diff mbox

Patch

Index: gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/config/rs6000/rs6000.c	(revision 209205)
+++ gcc/config/rs6000/rs6000.c	(working copy)
@@ -5635,11 +5635,15 @@  rs6000_expand_vector_set (rtx target, rtx val, int
 			UNSPEC_VPERM);
   else 
     {
-      /* Invert selector.  */
+      /* Invert selector.  We prefer to generate VNAND on P8 so
+         that future fusion opportunities can kick in, but must
+         generate VNOR elsewhere.  */
       rtx notx = gen_rtx_NOT (V16QImode, force_reg (V16QImode, x));
-      rtx andx = gen_rtx_AND (V16QImode, notx, notx);
+      rtx iorx = (TARGET_P8_VECTOR
+		  ? gen_rtx_IOR (V16QImode, notx, notx)
+		  : gen_rtx_AND (V16QImode, notx, notx));
       rtx tmp = gen_reg_rtx (V16QImode);
-      emit_move_insn (tmp, andx);
+      emit_insn (gen_rtx_SET (VOIDmode, tmp, iorx));
 
       /* Permute with operands reversed and adjusted selector.  */
       x = gen_rtx_UNSPEC (mode, gen_rtvec (3, reg, target, tmp),
@@ -30212,12 +30216,12 @@  altivec_expand_vec_perm_const_le (rtx operands[4])
 
 /* Similarly to altivec_expand_vec_perm_const_le, we must adjust the
    permute control vector.  But here it's not a constant, so we must
-   generate a vector NOR to do the adjustment.  */
+   generate a vector NAND or NOR to do the adjustment.  */
 
 void
 altivec_expand_vec_perm_le (rtx operands[4])
 {
-  rtx notx, andx, unspec;
+  rtx notx, iorx, unspec;
   rtx target = operands[0];
   rtx op0 = operands[1];
   rtx op1 = operands[2];
@@ -30236,10 +30240,13 @@  altivec_expand_vec_perm_le (rtx operands[4])
   if (!REG_P (target))
     tmp = gen_reg_rtx (mode);
 
-  /* Invert the selector with a VNOR.  */
+  /* Invert the selector with a VNAND if available, else a VNOR.
+     The VNAND is preferred for future fusion opportunities.  */
   notx = gen_rtx_NOT (V16QImode, sel);
-  andx = gen_rtx_AND (V16QImode, notx, notx);
-  emit_move_insn (norreg, andx);
+  iorx = (TARGET_P8_VECTOR
+	  ? gen_rtx_IOR (V16QImode, notx, notx)
+	  : gen_rtx_AND (V16QImode, notx, notx));
+  emit_insn (gen_rtx_SET (VOIDmode, norreg, iorx));
 
   /* Permute with operands reversed and adjusted selector.  */
   unspec = gen_rtx_UNSPEC (mode, gen_rtvec (3, op1, op0, norreg),