cmd/compile/internal/ssa: use BEQ/BNE to optimize the combination of XOR and EQ/NE on loong64

Reduce the number of go toolchain instructions on loong64 as follows:

	file	    before	after	     Δ		 %
	go	    1599056	1590560	    -8496    -0.5313%
	gofmt	    326188	326104	    -84	     -0.0258%
	asm	    563482	561250	    -2232    -0.3961%
	cgo	    488644	485252	    -3392    -0.6942%
	compile	    2504614	2486388	    -18226   -0.7277%
	cover	    526322	523270	    -3052    -0.5799%
	link	    714532	711124	    -3408    -0.4770%
	preprofile  242316	241112	    -1204    -0.4969%
	vet	    794446	786118	    -8328    -1.0483%

Change-Id: I0914889119a28ea672b694529ef54513fbb3f3b5
Reviewed-on: https://go-review.googlesource.com/c/go/+/693875
LUCI-TryBot-Result: Go LUCI <[email protected]>
Reviewed-by: abner chenc <[email protected]>
Reviewed-by: Dmitri Shuralyov <[email protected]>
Reviewed-by: Keith Randall <[email protected]>

Author: limeidan

src/cmd/compile/internal/ssa/_gen/LOONG64latelower.rules

@@ -4,3 +4,6 @@
 
 // Prefer addition when shifting left by one.
 (SLLVconst [1] x) => (ADDV x x)
+
+(EQZ (XOR x y) yes no) => (BEQ x y yes no)
+(NEZ (XOR x y) yes no) => (BNE x y yes no)