Use word count with word optimized block transfers

SPI.h

 
 	void inline transfer(void *buf, size_t count) {
 #if 0
-		// Ethernet library fails if this is used.  Why?
+		// TODO: byte order still needs work to match SPISettings
 		if (__builtin_constant_p(count)) {
 			if (count < 1) return;
 			if (((count & 3) == 0) && (((uint32_t)buf & 3) == 0)) {
 				// size is multiple of 4 and buffer is 32 bit aligned
-				transfer32(buf, buf, count);
+				transfer32(buf, buf, count >> 2);
 				return;
 			}
 			if (((count & 1) == 0) && (((uint32_t)buf & 1) == 0)) {
 				// size is multiple of 2 and buffer is 16 bit aligned
-				transfer16(buf, buf, count);
+				transfer16(buf, buf, count >> 1);
 				return;
 			}
 		}