[dev.power64] 9g: fix under-zeroing in clearfat

All three cases of clearfat were wrong on power64x.

The cases that handle 1032 bytes and up and 32 bytes and up
both use MOVDU (one directly generated in a loop and the other
via duffzero), which leaves the pointer register pointing at
the *last written* address.  The generated code was not
accounting for this, so the byte fill loop was re-zeroing the
last zeroed dword, rather than the bytes following the last
zeroed dword.  Fix this by simply adding an additional 8 byte
offset to the byte zeroing loop.

The case that handled under 32 bytes was also wrong.  It
didn't update the pointer register at all, so the byte zeroing
loop was simply re-zeroing the beginning of region.  Again,
the fix is to add an offset to the byte zeroing loop to
account for this.

LGTM=dave, bradfitz
R=rsc, dave, bradfitz
CC=golang-codereviews
https://codereview.appspot.com/168870043

1 files changed, 13 insertions, 7 deletions

diff --git a/src/cmd/9g/ggen.c b/src/cmd/9g/ggen.c
index c41d8eb41..7d9cf5050 100644
--- a/src/cmd/9g/ggen.c
+++ b/src/cmd/9g/ggen.c
@@ -900,7 +900,7 @@ ret:
 void
 clearfat(Node *nl)
 {
-	uint64 w, c, q, t;
+	uint64 w, c, q, t, boff;
 	Node dst, end, r0, *f;
 	Prog *p, *pl;
 
@@ -944,6 +944,8 @@ clearfat(Node *nl)
 		patch(gbranch(ABNE, T, 0), pl);
 
 		regfree(&end);
+		// The loop leaves R3 on the last zeroed dword
+		boff = 8;
 	} else if(q >= 4) {
 		p = gins(ASUB, N, &dst);
 		p->from.type = D_CONST;
@@ -953,17 +955,21 @@ clearfat(Node *nl)
 		afunclit(&p->to, f);
 		// 4 and 128 = magic constants: see ../../runtime/asm_power64x.s
 		p->to.offset = 4*(128-q);
-	} else
-	for(t = 0; t < q; t++) {
-		p = gins(AMOVD, &r0, &dst);
-		p->to.type = D_OREG;
-		p->to.offset = 8*t;
+		// duffzero leaves R3 on the last zeroed dword
+		boff = 8;
+	} else {
+		for(t = 0; t < q; t++) {
+			p = gins(AMOVD, &r0, &dst);
+			p->to.type = D_OREG;
+			p->to.offset = 8*t;
+		}
+		boff = 8*q;
 	}
 
 	for(t = 0; t < c; t++) {
 		p = gins(AMOVB, &r0, &dst);
 		p->to.type = D_OREG;
-		p->to.offset = t;
+		p->to.offset = t+boff;
 	}
 	reg[REGRT1]--;
 }