diff --git a/src/cmd/compile/internal/ppc64/ggen.go b/src/cmd/compile/internal/ppc64/ggen.go
index 3ae6422bf9..7877be3336 100644
--- a/src/cmd/compile/internal/ppc64/ggen.go
+++ b/src/cmd/compile/internal/ppc64/ggen.go
@@ -46,10 +46,9 @@ func zerorange(pp *objw.Progs, p *obj.Prog, off, cnt int64, _ *uint32) *obj.Prog
 }
 
 func ginsnop(pp *objw.Progs) *obj.Prog {
+	// Generate the preferred hardware nop: ori 0,0,0
 	p := pp.Prog(ppc64.AOR)
-	p.From.Type = obj.TYPE_REG
-	p.From.Reg = ppc64.REG_R0
-	p.To.Type = obj.TYPE_REG
-	p.To.Reg = ppc64.REG_R0
+	p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: 0}
+	p.To = obj.Addr{Type: obj.TYPE_REG, Reg: ppc64.REG_R0}
 	return p
 }
diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go
index 31fbb7f7bf..70ce9050b6 100644
--- a/src/cmd/internal/obj/ppc64/asm9.go
+++ b/src/cmd/internal/obj/ppc64/asm9.go
@@ -2552,7 +2552,13 @@ func (c *ctxt9) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		case AROTLW:
 			o1 = OP_RLW(OP_RLWNM, uint32(p.To.Reg), uint32(r), uint32(p.From.Reg), 0, 31)
 		default:
-			o1 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), uint32(p.From.Reg))
+			if p.As == AOR && p.From.Type == obj.TYPE_CONST && p.From.Offset == 0 {
+				// Compile "OR $0, Rx, Ry" into ori. If Rx == Ry == 0, this is the preferred
+				// hardware no-op. This happens because $0 matches C_REG before C_ZCON.
+				o1 = LOP_IRR(OP_ORI, uint32(p.To.Reg), uint32(r), 0)
+			} else {
+				o1 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), uint32(p.From.Reg))
+			}
 		}
 
 	case 7: /* mov r, soreg ==> stw o(r) */