diff --git a/src/runtime/memclr_ppc64x.s b/src/runtime/memclr_ppc64x.s
index 442faa25f2..f7375dbee6 100644
--- a/src/runtime/memclr_ppc64x.s
+++ b/src/runtime/memclr_ppc64x.s
@@ -7,25 +7,56 @@
 #include "textflag.h"
 
 // void runtime·memclr(void*, uintptr)
-TEXT runtime·memclr(SB),NOSPLIT|NOFRAME,$0-16
-	MOVD	ptr+0(FP), R3
-	MOVD	n+8(FP), R4
-	SRADCC	$3, R4, R6	// R6 is the number of words to zero
-	BEQ	bytes
+TEXT runtime·memclr(SB), NOSPLIT|NOFRAME, $0-16
+	MOVD ptr+0(FP), R3
+	MOVD n+8(FP), R4
 
-	SUB	$8, R3
-	MOVD	R6, CTR
-	MOVDU	R0, 8(R3)
-	BC	25, 0, -1(PC)	// bdnz+ $-4
-	ADD	$8, R3
+	// Determine if there are doublewords to clear
+check:
+	ANDCC $7, R4, R5  // R5: leftover bytes to clear
+	SRAD  $3, R4, R6  // R6: double words to clear
+	CMP   R6, $0, CR1 // CR1[EQ] set if no double words
 
-bytes:
-	ANDCC	$7, R4, R7	// R7 is the number of bytes to zero
-	BEQ	done
-	SUB	$1, R3
-	MOVD	R7, CTR
-	MOVBU	R0, 1(R3)
-	BC	25, 0, -1(PC)	// bdnz+ $-4
+	BC     12, 6, nozerolarge // only single bytes
+	MOVD   R6, CTR            // R6 = number of double words
+	SRADCC $2, R6, R7         // 32 byte chunks?
+	BNE    zero32setup
 
-done:
+	// Clear double words
+
+zero8:
+	MOVD R0, 0(R3)    // double word
+	ADD  $8, R3
+	BC   16, 0, zero8 // dec ctr, br zero8 if ctr not 0
+	BR   nozerolarge  // handle remainder
+
+	// Prepare to clear 32 bytes at a time.
+
+zero32setup:
+	DCBTST (R3)    // prepare data cache
+	MOVD   R7, CTR // number of 32 byte chunks
+
+zero32:
+	MOVD    R0, 0(R3)       // clear 4 double words
+	MOVD    R0, 8(R3)
+	MOVD    R0, 16(R3)
+	MOVD    R0, 24(R3)
+	ADD     $32, R3
+	BC      16, 0, zero32   // dec ctr, br zero32 if ctr not 0
+	RLDCLCC $61, R4, $3, R6 // remaining doublewords
+	BEQ     nozerolarge
+	MOVD    R6, CTR         // set up the CTR for doublewords
+	BR      zero8
+
+nozerolarge:
+	CMP R5, $0   // any remaining bytes
+	BC  4, 1, LR // ble lr
+
+zerotail:
+	MOVD R5, CTR // set up to clear tail bytes
+
+zerotailloop:
+	MOVB R0, 0(R3)           // clear single bytes
+	ADD  $1, R3
+	BC   16, 0, zerotailloop // dec ctr, br zerotailloop if ctr not 0
 	RET