internal/subtle: optimize amd64

internal/subtle/xor_amd64.s

@@ -24,6 +24,7 @@ non_avx2:
 aligned:
 	MOVQ $0, AX // position in slices
 
+	PCALIGN $16
 loop16b:
 	MOVOU (SI)(AX*1), X0   // XOR 16byte forwards.
 	MOVOU (CX)(AX*1), X1
@@ -34,6 +35,7 @@ loop16b:
 	JNE   loop16b
 	RET
 
+	PCALIGN $16
 loop_1b:
 	SUBQ  $1, DX           // XOR 1byte backwards.
 	MOVB  (SI)(DX*1), DI
@@ -62,34 +64,34 @@ ret:
 	RET
 
 avx2:
-	TESTQ $31, DX            // AND 31 & len, if not zero jump to avx2_not_aligned.
+	TESTQ $31, DX          // AND 31 & len, if not zero jump to avx2_not_aligned.
 	JNZ   avx2_not_aligned
 
-avx2_aligned:
+avx2_aligned:              // input length = 16*n, where n is greater or equal 2.
 	TESTQ $16, DX          // AND 16 & len, if zero jump to loop32b_start.
 	JE loop32b_start
 	SUBQ  $16, DX          // XOR 16bytes backwards.
-	MOVOU (SI)(DX*1), X0
-	MOVOU (CX)(DX*1), X1
-	PXOR  X1, X0
-	MOVOU X0, (BX)(DX*1)
-	CMPQ  DX, $0           // if len is 0, ret.
-	JE avx2_ret
+	VMOVDQU (SI)(DX*1), X0
+	VPXOR  (CX)(DX*1), X0, X0
+	VMOVDQU X0, (BX)(DX*1)
 
 loop32b_start:
 	MOVQ $0, AX            // position in slices
 
+	PCALIGN $32
 loop32b:
 	VMOVDQU (SI)(AX*1), Y0   // XOR 32byte forwards.
-	VMOVDQU (CX)(AX*1), Y1
-	VPXOR Y0, Y1, Y0
+	VPXOR (CX)(AX*1), Y0, Y0
 	VMOVDQU Y0, (BX)(AX*1)
 	ADDQ  $32, AX
 	CMPQ  DX, AX
 	JNE   loop32b
+
+avx2_ret:	
 	VZEROUPPER
 	RET
 
+	PCALIGN $16
 avx2_loop_1b:
 	SUBQ  $1, DX           // XOR 1byte backwards.
 	MOVB  (SI)(DX*1), DI
@@ -98,25 +100,17 @@ avx2_loop_1b:
 	MOVB  DI, (BX)(DX*1)
 	TESTQ $7, DX           // AND 7 & len, if not zero jump to avx2_loop_1b.
 	JNZ   avx2_loop_1b
-	CMPQ  DX, $0           // if len is 0, ret.
-	JE    avx2_ret
 	TESTQ $15, DX          // AND 15 & len, if zero jump to aligned.
 	JZ    avx2_aligned
 
 avx2_not_aligned:
 	TESTQ $7, DX           // AND $7 & len, if not zero jump to avx2_loop_1b.
 	JNE   avx2_loop_1b
-	TESTQ $8, DX           // AND $8 & len, if zero jump to avx2_16b.
-	JE   avx2_16b
+	TESTQ $8, DX           // AND $8 & len, if zero jump to avx2_aligned.
+	JE   avx2_aligned
 	SUBQ  $8, DX           // XOR 8bytes backwards.
 	MOVQ  (SI)(DX*1), DI
 	MOVQ  (CX)(DX*1), AX
 	XORQ  AX, DI
 	MOVQ  DI, (BX)(DX*1)
-avx2_16b:	
-	CMPQ  DX, $16          // if len is greater or equal 16 here, it must be aligned.
-	JGE   avx2_aligned
-
-avx2_ret:
-	VZEROUPPER
-	RET
+	JMP  avx2_aligned