308 #  define block_copy_nn(d, s, l)    memcpy(d, s, l)  argument
309 #  define block_copy(d, s)          memcpy(d, s, N_BLOCK)  argument
311 #  define block_copy_nn(d, s, l)    copy_block_nn(d, s, l)  argument
312 #  define block_copy(d, s)          copy_block(d, s)  argument
315 static void copy_block( void *d, const void *s )  in copy_block()  argument
318     ((uint32_t*)d)[ 0] = ((uint32_t*)s)[ 0];  in copy_block()
319     ((uint32_t*)d)[ 1] = ((uint32_t*)s)[ 1];  in copy_block()
320     ((uint32_t*)d)[ 2] = ((uint32_t*)s)[ 2];  in copy_block()
321     ((uint32_t*)d)[ 3] = ((uint32_t*)s)[ 3];  in copy_block()
323     ((uint8_t*)d)[ 0] = ((uint8_t*)s)[ 0];  in copy_block()
324     ((uint8_t*)d)[ 1] = ((uint8_t*)s)[ 1];  in copy_block()
325     ((uint8_t*)d)[ 2] = ((uint8_t*)s)[ 2];  in copy_block()
326     ((uint8_t*)d)[ 3] = ((uint8_t*)s)[ 3];  in copy_block()
327     ((uint8_t*)d)[ 4] = ((uint8_t*)s)[ 4];  in copy_block()
328     ((uint8_t*)d)[ 5] = ((uint8_t*)s)[ 5];  in copy_block()
329     ((uint8_t*)d)[ 6] = ((uint8_t*)s)[ 6];  in copy_block()
330     ((uint8_t*)d)[ 7] = ((uint8_t*)s)[ 7];  in copy_block()
331     ((uint8_t*)d)[ 8] = ((uint8_t*)s)[ 8];  in copy_block()
332     ((uint8_t*)d)[ 9] = ((uint8_t*)s)[ 9];  in copy_block()
333     ((uint8_t*)d)[10] = ((uint8_t*)s)[10];  in copy_block()
334     ((uint8_t*)d)[11] = ((uint8_t*)s)[11];  in copy_block()
335     ((uint8_t*)d)[12] = ((uint8_t*)s)[12];  in copy_block()
336     ((uint8_t*)d)[13] = ((uint8_t*)s)[13];  in copy_block()
337     ((uint8_t*)d)[14] = ((uint8_t*)s)[14];  in copy_block()
338     ((uint8_t*)d)[15] = ((uint8_t*)s)[15];  in copy_block()
342 static void copy_block_nn( uint8_t * d, const uint8_t *s, uint8_t nn )  in copy_block_nn()  argument
346         *d++ = *s++;  in copy_block_nn()
349 static void xor_block( void *d, const void *s )  in xor_block()  argument
352     ((uint32_t*)d)[ 0] ^= ((uint32_t*)s)[ 0];  in xor_block()
353     ((uint32_t*)d)[ 1] ^= ((uint32_t*)s)[ 1];  in xor_block()
354     ((uint32_t*)d)[ 2] ^= ((uint32_t*)s)[ 2];  in xor_block()
355     ((uint32_t*)d)[ 3] ^= ((uint32_t*)s)[ 3];  in xor_block()
357     ((uint8_t*)d)[ 0] ^= ((uint8_t*)s)[ 0];  in xor_block()
358     ((uint8_t*)d)[ 1] ^= ((uint8_t*)s)[ 1];  in xor_block()
359     ((uint8_t*)d)[ 2] ^= ((uint8_t*)s)[ 2];  in xor_block()
360     ((uint8_t*)d)[ 3] ^= ((uint8_t*)s)[ 3];  in xor_block()
361     ((uint8_t*)d)[ 4] ^= ((uint8_t*)s)[ 4];  in xor_block()
362     ((uint8_t*)d)[ 5] ^= ((uint8_t*)s)[ 5];  in xor_block()
363     ((uint8_t*)d)[ 6] ^= ((uint8_t*)s)[ 6];  in xor_block()
364     ((uint8_t*)d)[ 7] ^= ((uint8_t*)s)[ 7];  in xor_block()
365     ((uint8_t*)d)[ 8] ^= ((uint8_t*)s)[ 8];  in xor_block()
366     ((uint8_t*)d)[ 9] ^= ((uint8_t*)s)[ 9];  in xor_block()
367     ((uint8_t*)d)[10] ^= ((uint8_t*)s)[10];  in xor_block()
368     ((uint8_t*)d)[11] ^= ((uint8_t*)s)[11];  in xor_block()
369     ((uint8_t*)d)[12] ^= ((uint8_t*)s)[12];  in xor_block()
370     ((uint8_t*)d)[13] ^= ((uint8_t*)s)[13];  in xor_block()
371     ((uint8_t*)d)[14] ^= ((uint8_t*)s)[14];  in xor_block()
372     ((uint8_t*)d)[15] ^= ((uint8_t*)s)[15];  in xor_block()
376 static void copy_and_key( void *d, const void *s, const void *k )  in copy_and_key()  argument
379     ((uint32_t*)d)[ 0] = ((uint32_t*)s)[ 0] ^ ((uint32_t*)k)[ 0];  in copy_and_key()
380     ((uint32_t*)d)[ 1] = ((uint32_t*)s)[ 1] ^ ((uint32_t*)k)[ 1];  in copy_and_key()
381     ((uint32_t*)d)[ 2] = ((uint32_t*)s)[ 2] ^ ((uint32_t*)k)[ 2];  in copy_and_key()
382     ((uint32_t*)d)[ 3] = ((uint32_t*)s)[ 3] ^ ((uint32_t*)k)[ 3];  in copy_and_key()
384     ((uint8_t*)d)[ 0] = ((uint8_t*)s)[ 0] ^ ((uint8_t*)k)[ 0];  in copy_and_key()
385     ((uint8_t*)d)[ 1] = ((uint8_t*)s)[ 1] ^ ((uint8_t*)k)[ 1];  in copy_and_key()
386     ((uint8_t*)d)[ 2] = ((uint8_t*)s)[ 2] ^ ((uint8_t*)k)[ 2];  in copy_and_key()
387     ((uint8_t*)d)[ 3] = ((uint8_t*)s)[ 3] ^ ((uint8_t*)k)[ 3];  in copy_and_key()
388     ((uint8_t*)d)[ 4] = ((uint8_t*)s)[ 4] ^ ((uint8_t*)k)[ 4];  in copy_and_key()
389     ((uint8_t*)d)[ 5] = ((uint8_t*)s)[ 5] ^ ((uint8_t*)k)[ 5];  in copy_and_key()
390     ((uint8_t*)d)[ 6] = ((uint8_t*)s)[ 6] ^ ((uint8_t*)k)[ 6];  in copy_and_key()
391     ((uint8_t*)d)[ 7] = ((uint8_t*)s)[ 7] ^ ((uint8_t*)k)[ 7];  in copy_and_key()
392     ((uint8_t*)d)[ 8] = ((uint8_t*)s)[ 8] ^ ((uint8_t*)k)[ 8];  in copy_and_key()
393     ((uint8_t*)d)[ 9] = ((uint8_t*)s)[ 9] ^ ((uint8_t*)k)[ 9];  in copy_and_key()
394     ((uint8_t*)d)[10] = ((uint8_t*)s)[10] ^ ((uint8_t*)k)[10];  in copy_and_key()
395     ((uint8_t*)d)[11] = ((uint8_t*)s)[11] ^ ((uint8_t*)k)[11];  in copy_and_key()
396     ((uint8_t*)d)[12] = ((uint8_t*)s)[12] ^ ((uint8_t*)k)[12];  in copy_and_key()
397     ((uint8_t*)d)[13] = ((uint8_t*)s)[13] ^ ((uint8_t*)k)[13];  in copy_and_key()
398     ((uint8_t*)d)[14] = ((uint8_t*)s)[14] ^ ((uint8_t*)k)[14];  in copy_and_key()
399     ((uint8_t*)d)[15] = ((uint8_t*)s)[15] ^ ((uint8_t*)k)[15];  in copy_and_key()
401     block_copy(d, s);  in copy_and_key()
402     xor_block(d, k);  in copy_and_key()
406 static void add_round_key( uint8_t d[N_BLOCK], const uint8_t k[N_BLOCK] )  in add_round_key()
408     xor_block(d, k);  in add_round_key()