internal/subtle: apply VLM/VSTM in non-loop

emmansun · Aug 28, 2024 · 45584cc · 45584cc
1 parent 40bd384
commit 45584cc
Showing 1 changed file with 18 additions and 9 deletions.
diff --git a/internal/subtle/xor_s390x.s b/internal/subtle/xor_s390x.s
@@ -13,28 +13,37 @@ TEXT ·xorBytes(SB),NOSPLIT,$0-32
 	CMPBLT	R4, $64, tail
 
 loop_64:
-	VLM 0(R2)(R5*1), V0, V3
-	VLM 0(R3)(R5*1), V4, V7
+	VL 0(R2)(R5*1), V0
+	VL 16(R2)(R5*1), V1
+	VL 32(R2)(R5*1), V2
+	VL 48(R2)(R5*1), V3
+	VL 0(R3)(R5*1), V4
+	VL 16(R3)(R5*1), V5
+	VL 32(R3)(R5*1), V6
+	VL 48(R3)(R5*1), V7
 	VX V0, V4, V4
 	VX V1, V5, V5
 	VX V2, V6, V6
 	VX V3, V7, V7
-	VSTM V4, V7, 0(R1)(R5*1)
+	VST V4, 0(R1)(R5*1)
+	VST V5, 16(R1)(R5*1)
+	VST V6, 32(R1)(R5*1)
+	VST V7, 48(R1)(R5*1)
 	LAY	64(R5), R5
 	SUB	$64, R4
 	CMPBGE	R4, $64, loop_64
 
 tail:
 	CMPBEQ	R4, $0, done
 	CMPBLT	R4, $32, less_than32
-	VL 0(R2)(R5*1), V0
-	VL 16(R2)(R5*1), V1
-	VL 0(R3)(R5*1), V2
-	VL 16(R3)(R5*1), V3
+	VLM 0(R2)(R5*1), V0, V1
+	//VL 16(R2)(R5*1), V1
+	VLM 0(R3)(R5*1), V2, V3
+	//VL 16(R3)(R5*1), V3
 	VX V0, V2, V2
 	VX V1, V3, V3
-	VST V2, 0(R1)(R5*1)
-	VST V3, 16(R1)(R5*1)
+	VSTM V2, V3, 0(R1)(R5*1)
+	//VST V3, 16(R1)(R5*1)
 	LAY	32(R5), R5
 	SUB	$32, R4