Optimize the rary_splice().

* before
               user     system      total        real
len == rlen  0.860000   0.000000   0.860000 (  0.852845)
len != rlen 21.180000   0.010000  21.190000 ( 21.186979)

* after
                user     system      total        real
len == rlen  0.050000   0.000000   0.050000 (  0.056620)
len != rlen  0.150000   0.000000   0.150000 (  0.145373)

{{{
require 'benchmark'

Benchmark.bm(10) do |x|
  ary =  Array.new(1000)
  tmp =  ["a"] * 1100
  x.report "len ==  rlen" do
    10000.times do
      ary[100, tmp.size] =  tmp
    end
  end

  x.report "len !=  rlen" do
    10000.times do
      ary[100, tmp.size - 1] =  tmp
    end
  end
end
}}}

array.c

@@ -942,14 +942,36 @@ rary_splice(VALUE ary, long beg, long len, VALUE rpl)
 	}
     }
     else if (len == rlen) {
-	for (long i = 0; i < len; i++) {
-	    rary_elt_set(ary, beg + i, rb_ary_elt(rpl, i));
-	}	
+	if (rlen > 0 && IS_RARY(rpl)) {
+	    GC_MEMMOVE(&RARY(ary)->elements[RARY(ary)->beg + beg],
+		       &RARY(rpl)->elements[RARY(rpl)->beg],
+		       sizeof(VALUE) * len);
+	}
+	else {
+	    for (long i = 0; i < len; i++) {
+		rary_elt_set(ary, beg + i, rb_ary_elt(rpl, i));
+	    }
+	}
     }
     else {
-	rary_erase(ary, beg, len);
-	for (long i = 0; i < rlen; i++) {
-	    rary_insert(ary, beg + i, rb_ary_elt(rpl, i));
+	if (rlen > 0 && IS_RARY(rpl)) {
+	    long newlen = RARY(ary)->len + rlen - len;
+
+	    rary_reserve(ary, newlen);
+	    GC_MEMMOVE(&RARY(ary)->elements[RARY(ary)->beg + beg + rlen],
+		       &RARY(ary)->elements[RARY(ary)->beg + beg + len],
+		       sizeof(VALUE) * (RARY(ary)->len - (beg + len)));
+
+	    GC_MEMMOVE(&RARY(ary)->elements[RARY(ary)->beg + beg],
+		       rary_ptr(rpl),
+		       sizeof(VALUE) * rlen);
+	    RARY(ary)->len = newlen;
+	}
+	else {
+	    rary_erase(ary, beg, len);
+	    for (long i = 0; i < rlen; i++) {
+		rary_insert(ary, beg + i, rb_ary_elt(rpl, i));
+	    }
 	}
     }
 }