+serpent bitslice; modifyed cast5 testings

diff --git a/cast5.mk b/cast5.mk
index f2241c0d8e4e5042f0e2c6808c66f300898981cf..23dadc116c461f16d9f0de7be82d23059e4b717e 100644 (file)
--- a/cast5.mk
+++ b/cast5.mk
@@ -5,7 +5,8 @@ ALGO_NAME := CAST5
 BLOCK_CIPHERS += $(ALGO_NAME)
 
 $(ALGO_NAME)_OBJ      := cast5.o
-$(ALGO_NAME)_TEST_BIN := main-cast5-test.o debug.o uart.o serial-tools.o cast5.o
+$(ALGO_NAME)_TEST_BIN := main-cast5-test.o debug.o uart.o serial-tools.o \
+                         cast5.o nessie_bc_test.o
 $(ALGO_NAME)_NESSIE_TEST      := "nessie"
 $(ALGO_NAME)_PEROFRMANCE_TEST := "performance"
 

diff --git a/main-cast5-test.c b/main-cast5-test.c
index 9b3795a39d2608769a9603a1b8431b8524bfcc3a..6f3b18a71228afd9061c4d1f75e72448226f21cf 100644 (file)
--- a/main-cast5-test.c
+++ b/main-cast5-test.c
@@ -9,14 +9,45 @@
 #include "debug.h"
 
 #include "cast5.h"
+#include "nessie_bc_test.h"
 
 #include <stdint.h>
 #include <string.h>
 
+char* cipher_name = "cast-128 (cast5)";
 
 /*****************************************************************************
  *  additional validation-functions                                                                                     *
  *****************************************************************************/
+/*
+       void cast5_init(cast5_ctx_t* s, uint8_t* key, uint8_t keylength);
+       void cast5_enc(cast5_ctx_t *s, void* block);
+       void cast5_dec(cast5_ctx_t *s, void* block);
+*/ 
+
+void cast5_init_dummy(uint8_t* key, uint8_t keylength_b, cast5_ctx_t* ctx){
+       cast5_init(ctx, key, keylength_b);
+}
+
+void cast5_enc_dummy(void* buffer, cast5_ctx_t* ctx){
+       cast5_enc(ctx, buffer);
+}
+
+void cast5_dec_dummy(void* buffer, cast5_ctx_t* ctx){
+       cast5_dec(ctx, buffer);
+}
+
+void test_nessie_cast5(void){
+       nessie_bc_ctx.blocksize_B =   8;
+       nessie_bc_ctx.keysize_b   = 128;
+       nessie_bc_ctx.name        = cipher_name;
+       nessie_bc_ctx.ctx_size_B  = sizeof(cast5_ctx_t);
+       nessie_bc_ctx.cipher_enc  = (nessie_bc_enc_fpt)cast5_enc_dummy;
+       nessie_bc_ctx.cipher_dec  = (nessie_bc_dec_fpt)cast5_dec_dummy;
+       nessie_bc_ctx.cipher_genctx  = (nessie_bc_gen_fpt)cast5_init_dummy;
+       
+       nessie_bc_run();
+}
 
 /*****************************************************************************
  *  self tests                                                                                                                          *
@@ -135,8 +166,9 @@ int main (void){
 restart:
        while(1){ 
                if (!getnextwordn(str,20))  {DEBUG_S("DBG: W1\r\n"); goto error;}
-               if (strcmp(str, "test")) {DEBUG_S("DBG: 1b\r\n"); goto error;}
-                       testrun_cast5();
+               if (strcmp(str, "nessie")) {DEBUG_S("DBG: 1b\r\n"); goto error;}
+               //      testrun_cast5();
+                       test_nessie_cast5();
                goto restart;           
                continue;
        error:

diff --git a/nessie_bc_test.c b/nessie_bc_test.c
index 4d6d42b38c09bca42452afbae23428cd1515822a..c8f925738af7b964a8e99f1b4259ef9419e362d2 100644 (file)
--- a/nessie_bc_test.c
+++ b/nessie_bc_test.c
@@ -241,9 +241,10 @@ void nessie_bc_run(void){
        for(i=0; i<nessie_bc_ctx.blocksize_B; ++i){
                buffer[i]=kasumi_plain[i%sizeof(kasumi_plain)];
        }
+       nessie_bc_enc(key, buffer);
        /* half done ;-) */
        /* test set 5 */
-       set=1;
+       set=5;
        print_setheader(set);
        for(i=0; i<nessie_bc_ctx.keysize_b; ++i){
                print_set_vector(set, i);
@@ -291,5 +292,6 @@ void nessie_bc_run(void){
        for(i=0; i<nessie_bc_ctx.blocksize_B; ++i){
                buffer[i]=kasumi_plain[i%sizeof(kasumi_plain)];
        }
+       nessie_bc_dec(key, buffer);
        print_footer();
 }

diff --git a/serpent-sboxes-bitslice.c b/serpent-sboxes-bitslice.c
new file mode 100644 (file)
index 0000000..47a22fb
--- /dev/null
+++ b/serpent-sboxes-bitslice.c
@@ -0,0 +1,461 @@
+/* serpent-sboxes.c
+ * a non-bitsliced implementation of the serpent sboxes
+ * author: Daniel Otte 
+ * email:  daniel.otte@rub.de
+ * license: GPLv3
+ */
+
+#include <stdint.h>
+#include <string.h>
+#include <avr/pgmspace.h>
+#include "serpent-sboxes.h"
+
+
+
+
+/* S0:   3  8 15  1 10  6  5 11 14 13  4  2  7  0  9 12 */
+
+/* depth = 5,7,4,2, Total gates=18 */
+static
+void sb0(uint8_t* out, const uint8_t* in, uint8_t i){
+// (a,b,c,d,w,x,y,z) 
+       uint8_t t01, t02, t03, t05, t06, t07, t08, t09, t11, t12, t13, t14, t15, t17;
+       t01          = in[4*1+i]  ^ in[4*2+i]; 
+       t02          = in[4*0+i]  | in[4*3+i]; 
+       t03          = in[4*0+i]  ^ in[4*1+i]; 
+       out[4*3+i]   = t02        ^ t01; 
+       t05          = in[4*2+i]  | out[4*3+i]; 
+       t06          = in[4*0+i]  ^ in[4*3+i]; 
+       t07          = in[4*1+i]  | in[4*2+i]; 
+       t08          = in[4*3+i]  & t05; 
+       t09          = t03        & t07; 
+       out[4*2+i]   = t09        ^ t08; 
+       t11          = t09        & out[4*2+i]; 
+       t12          = in[4*2+i]  ^ in[4*3+i]; 
+       t13          = t07        ^ t11; 
+       t14          = in[4*1+i]  & t06; 
+       t15          = t06        ^ t13; 
+       out[4*0+i]   =     ~ t15; 
+       t17          = out[4*0+i] ^ t14; 
+       out[4*1+i]   = t12        ^ t17; 
+}
+
+/* InvS0:  13  3 11  0 10  6  5 12  1 14  4  7 15  9  8  2 */
+
+/* depth = 8,4,3,6, Total gates=19 */
+static
+void sb0_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t08, t09, t10, t12, t13, t14, t15, t17, t18, t01;
+       t01          = in[4*2+i] ^ in[4*3+i];
+       t02          = in[4*0+i] | in[4*1+i];
+       t03          = in[4*1+i] | in[4*2+i];
+       t04          = in[4*2+i] & t01;
+       t05          = t02       ^ t01;
+       t06          = in[4*0+i] | t04;
+       out[4*2+i]   =     ~ t05;
+       t08          = in[4*1+i]   ^ in[4*3+i];
+       t09          = t03       & t08;
+       t10          = in[4*3+i]   | out[4*2+i];
+       out[4*1+i]   = t09       ^ t06;
+       t12          = in[4*0+i]   | t05;
+       t13          = out[4*1+i]   ^ t12;
+       t14          = t03       ^ t10;
+       t15          = in[4*0+i]   ^ in[4*2+i];
+       out[4*3+i]   = t14       ^ t13;
+       t17          = t05       & t13;
+       t18          = t14       | t17;
+       out[4*0+i]   = t15       ^ t18; 
+}
+
+/* S1:  15 12  2  7  9  0  5 10  1 11 14  8  6 13  3  4 */
+
+/* depth = 10,7,3,5, Total gates=18 */
+static 
+void sb1(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t08, t10, t11, t12, t13, t16, t17, t01;
+       t01          = in[4*0+i]   | in[4*3+i];
+       t02          = in[4*2+i]   ^ in[4*3+i];
+       t03          =     ~ in[4*1+i];
+       t04          = in[4*0+i]   ^ in[4*2+i];
+       t05          = in[4*0+i]   | t03;
+       t06          = in[4*3+i]   & t04;
+       t07          = t01       & t02;
+       t08          = in[4*1+i]   | t06;
+       out[4*2+i]   = t02       ^ t05;
+       t10          = t07       ^ t08;
+       t11          = t01       ^ t10;
+       t12          = out[4*2+i]   ^ t11;
+       t13          = in[4*1+i]   & in[4*3+i];
+       out[4*3+i]   =     ~ t10;
+       out[4*1+i]   = t13       ^ t12;
+       t16          = t10       | out[4*1+i];
+       t17          = t05       & t16;
+       out[4*0+i]   = in[4*2+i]   ^ t17; 
+}
+
+/* InvS1:   5  8  2 14 15  6 12  3 11  4  7  9  1 13 10  0 */
+
+/* depth = 7,4,5,3, Total gates=18 */
+static void sb1_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t08, t09, t10, t11, t14, t15, t17, t01;
+       t01          = in[4*0+i]   ^ in[4*1+i];
+       t02          = in[4*1+i]   | in[4*3+i];
+       t03          = in[4*0+i]   & in[4*2+i];
+       t04          = in[4*2+i]   ^ t02;
+       t05          = in[4*0+i]   | t04;
+       t06          = t01       & t05;
+       t07          = in[4*3+i]   | t03;
+       t08          = in[4*1+i]   ^ t06;
+       t09          = t07       ^ t06;
+       t10          = t04       | t03;
+       t11          = in[4*3+i]   & t08;
+       out[4*2+i]   =     ~ t09;
+       out[4*1+i]   = t10       ^ t11;
+       t14          = in[4*0+i]   | out[4*2+i];
+       t15          = t06       ^ out[4*1+i];
+       out[4*3+i]   = t01       ^ t04;
+       t17          = in[4*2+i]   ^ t15;
+       out[4*0+i]   = t14       ^ t17; 
+}
+
+/* S2:   8  6  7  9  3 12 10 15 13  1 14  4  0 11  5  2 */
+
+/* depth = 3,8,11,7, Total gates=16 */
+static void sb2(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t05, t06, t07, t08, t09, t10, t12, t13, t14, t01;
+       t01          = in[4*0+i]   | in[4*2+i];
+       t02          = in[4*0+i]   ^ in[4*1+i];
+       t03          = in[4*3+i]   ^ t01;
+       out[4*0+i]   = t02       ^ t03;
+       t05          = in[4*2+i]   ^ out[4*0+i];
+       t06          = in[4*1+i]   ^ t05;
+       t07          = in[4*1+i]   | t05;
+       t08          = t01       & t06;
+       t09          = t03       ^ t07;
+       t10          = t02       | t09;
+       out[4*1+i]   = t10       ^ t08;
+       t12          = in[4*0+i]   | in[4*3+i];
+       t13          = t09       ^ out[4*1+i];
+       t14          = in[4*1+i]   ^ t13;
+       out[4*3+i]   =     ~ t09;
+       out[4*2+i]   = t12       ^ t14; 
+}
+
+/* InvS2:  12  9 15  4 11 14  1  2  0  3  6 13  5  8 10  7 */
+
+/* depth = 3,6,8,3, Total gates=18 */
+static void sb2_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t06, t07, t08, t09, t10, t11, t12, t15, t16, t17, t01;
+       t01          = in[4*0+i]   ^ in[4*3+i];
+       t02          = in[4*2+i]   ^ in[4*3+i];
+       t03          = in[4*0+i]   & in[4*2+i];
+       t04          = in[4*1+i]   | t02;
+       out[4*0+i]   = t01       ^ t04;
+       t06          = in[4*0+i]   | in[4*2+i];
+       t07          = in[4*3+i]   | out[4*0+i];
+       t08          =     ~ in[4*3+i];
+       t09          = in[4*1+i]   & t06;
+       t10          = t08       | t03;
+       t11          = in[4*1+i]   & t07;
+       t12          = t06       & t02;
+       out[4*3+i]   = t09       ^ t10;
+       out[4*1+i]   = t12       ^ t11;
+       t15          = in[4*2+i]   & out[4*3+i];
+       t16          = out[4*0+i]   ^ out[4*1+i];
+       t17          = t10       ^ t15;
+       out[4*2+i]   = t16       ^ t17; 
+}
+
+/* S3:   0 15 11  8 12  9  6  3 13  1  2  4 10  7  5 14 */
+
+/* depth = 8,3,5,5, Total gates=18 */
+static void sb3(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t08, t09, t10, t11, t13, t14, t15, t01;
+       t01          = in[4*0+i]   ^ in[4*2+i];
+       t02          = in[4*0+i]   | in[4*3+i];
+       t03          = in[4*0+i]   & in[4*3+i];
+       t04          = t01       & t02;
+       t05          = in[4*1+i]   | t03;
+       t06          = in[4*0+i]   & in[4*1+i];
+       t07          = in[4*3+i]   ^ t04;
+       t08          = in[4*2+i]   | t06;
+       t09          = in[4*1+i]   ^ t07;
+       t10          = in[4*3+i]   & t05;
+       t11          = t02       ^ t10;
+       out[4*3+i]   = t08       ^ t09;
+       t13          = in[4*3+i]   | out[4*3+i];
+       t14          = in[4*0+i]   | t07;
+       t15          = in[4*1+i]   & t13;
+       out[4*2+i]   = t08       ^ t11;
+       out[4*0+i]   = t14       ^ t15;
+       out[4*1+i]   = t05       ^ t04; 
+}
+
+/* InvS3:   0  9 10  7 11 14  6 13  3  5 12  2  4  8 15  1 */
+
+/* depth = 3,6,4,4, Total gates=17 */
+static void sb3_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t09, t11, t12, t13, t14, t16, t01;
+       t01          = in[4*2+i]   | in[4*3+i];
+       t02          = in[4*0+i]   | in[4*3+i];
+       t03          = in[4*2+i]   ^ t02;
+       t04          = in[4*1+i]   ^ t02;
+       t05          = in[4*0+i]   ^ in[4*3+i];
+       t06          = t04       & t03;
+       t07          = in[4*1+i]   & t01;
+       out[4*2+i]   = t05       ^ t06;
+       t09          = in[4*0+i]   ^ t03;
+       out[4*0+i]   = t07       ^ t03;
+       t11          = out[4*0+i]   | t05;
+       t12          = t09       & t11;
+       t13          = in[4*0+i]   & out[4*2+i];
+       t14          = t01       ^ t05;
+       out[4*1+i]   = in[4*1+i]   ^ t12;
+       t16          = in[4*1+i]   | t13;
+       out[4*3+i]   = t14       ^ t16; 
+}
+
+/* S4:   1 15  8  3 12  0 11  6  2  5  4 10  9 14  7 13 */
+
+/* depth = 6,7,5,3, Total gates=19 */
+static void sb4(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t08, t09, t10, t11, t12, t13, t14, t15, t16, t01;
+       t01          = in[4*0+i]   | in[4*1+i];
+       t02          = in[4*1+i]   | in[4*2+i];
+       t03          = in[4*0+i]   ^ t02;
+       t04          = in[4*1+i]   ^ in[4*3+i];
+       t05          = in[4*3+i]   | t03;
+       t06          = in[4*3+i]   & t01;
+       out[4*3+i]   = t03       ^ t06;
+       t08          = out[4*3+i]   & t04;
+       t09          = t04       & t05;
+       t10          = in[4*2+i]   ^ t06;
+       t11          = in[4*1+i]   & in[4*2+i];
+       t12          = t04       ^ t08;
+       t13          = t11       | t03;
+       t14          = t10       ^ t09;
+       t15          = in[4*0+i]   & t05;
+       t16          = t11       | t12;
+       out[4*2+i]   = t13       ^ t08;
+       out[4*1+i]   = t15       ^ t16;
+       out[4*0+i]   =     ~ t14; 
+}
+
+/* InvS4:   5  0  8  3 10  9  7 14  2 12 11  6  4 15 13  1 */
+
+/* depth = 6,4,7,3, Total gates=17 */
+static void sb4_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t09, t10, t11, t12, t13, t15, t01;
+       t01          = in[4*1+i]   | in[4*3+i];
+       t02          = in[4*2+i]   | in[4*3+i];
+       t03          = in[4*0+i]   & t01;
+       t04          = in[4*1+i]   ^ t02;
+       t05          = in[4*2+i]   ^ in[4*3+i];
+       t06          =     ~ t03;
+       t07          = in[4*0+i]   & t04;
+       out[4*1+i]   = t05       ^ t07;
+       t09          = out[4*1+i]   | t06;
+       t10          = in[4*0+i]   ^ t07;
+       t11          = t01       ^ t09;
+       t12          = in[4*3+i]   ^ t04;
+       t13          = in[4*2+i]   | t10;
+       out[4*3+i]   = t03       ^ t12;
+       t15          = in[4*0+i]   ^ t04;
+       out[4*2+i]   = t11       ^ t13;
+       out[4*0+i]   = t15       ^ t09; 
+}
+
+/* S5:  15  5  2 11  4 10  9 12  0  3 14  8 13  6  7  1 */
+
+/* depth = 4,6,8,6, Total gates=17 */
+static void sb5(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t07, t08, t09, t10, t11, t12, t13, t14, t01;
+       t01          = in[4*1+i]   ^ in[4*3+i];
+       t02          = in[4*1+i]   | in[4*3+i];
+       t03          = in[4*0+i]   & t01;
+       t04          = in[4*2+i]   ^ t02;
+       t05          = t03       ^ t04;
+       out[4*0+i]   =     ~ t05;
+       t07          = in[4*0+i]   ^ t01;
+       t08          = in[4*3+i]   | out[4*0+i];
+       t09          = in[4*1+i]   | t05;
+       t10          = in[4*3+i]   ^ t08;
+       t11          = in[4*1+i]   | t07;
+       t12          = t03       | out[4*0+i];
+       t13          = t07       | t10;
+       t14          = t01       ^ t11;
+       out[4*2+i]   = t09       ^ t13;
+       out[4*1+i]   = t07       ^ t08;
+       out[4*3+i]   = t12       ^ t14; 
+}
+
+/* InvS5:   8 15  2  9  4  1 13 14 11  6  5  3  7 12 10  0 */
+
+/* depth = 4,6,9,7, Total gates=17 */
+static void sb5_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t07, t08, t09, t10, t12, t13, t15, t16, t01;
+       t01          = in[4*0+i]   & in[4*3+i];
+       t02          = in[4*2+i]   ^ t01;
+       t03          = in[4*0+i]   ^ in[4*3+i];
+       t04          = in[4*1+i]   & t02;
+       t05          = in[4*0+i]   & in[4*2+i];
+       out[4*0+i]   = t03       ^ t04;
+       t07          = in[4*0+i]   & out[4*0+i];
+       t08          = t01       ^ out[4*0+i];
+       t09          = in[4*1+i]   | t05;
+       t10          =     ~ in[4*1+i];
+       out[4*1+i]   = t08       ^ t09;
+       t12          = t10       | t07;
+       t13          = out[4*0+i]   | out[4*1+i];
+       out[4*3+i]   = t02       ^ t12;
+       t15          = t02       ^ t13;
+       t16          = in[4*1+i]   ^ in[4*3+i];
+       out[4*2+i]   = t16       ^ t15; 
+}
+
+/* S6:   7  2 12  5  8  4  6 11 14  9  1 15 13  3 10  0 */
+
+/* depth = 8,3,6,3, Total gates=19 */
+static void sb6(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t07, t08, t09, t10, t11, t12, t13, t15, t17, t18, t01;
+       t01          = in[4*0+i]   & in[4*3+i];
+       t02          = in[4*1+i]   ^ in[4*2+i];
+       t03          = in[4*0+i]   ^ in[4*3+i];
+       t04          = t01       ^ t02;
+       t05          = in[4*1+i]   | in[4*2+i];
+       out[4*1+i]   =     ~ t04;
+       t07          = t03       & t05;
+       t08          = in[4*1+i]   & out[4*1+i];
+       t09          = in[4*0+i]   | in[4*2+i];
+       t10          = t07       ^ t08;
+       t11          = in[4*1+i]   | in[4*3+i];
+       t12          = in[4*2+i]   ^ t11;
+       t13          = t09       ^ t10;
+       out[4*2+i]   =     ~ t13;
+       t15          = out[4*1+i]   & t03;
+       out[4*3+i]   = t12       ^ t07;
+       t17          = in[4*0+i]   ^ in[4*1+i];
+       t18          = out[4*2+i]   ^ t15;
+       out[4*0+i]   = t17       ^ t18; 
+}
+
+/* InvS6:  15 10  1 13  5  3  6  0  4  9 14  7  2 12  8 11 */
+
+/* depth = 5,3,8,6, Total gates=19 */
+static void sb6_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t07, t08, t09, t12, t13, t14, t15, t16, t17, t01;
+       t01          = in[4*0+i]   ^ in[4*2+i];
+       t02          =     ~ in[4*2+i];
+       t03          = in[4*1+i]   & t01;
+       t04          = in[4*1+i]   | t02;
+       t05          = in[4*3+i]   | t03;
+       t06          = in[4*1+i]   ^ in[4*3+i];
+       t07          = in[4*0+i]   & t04;
+       t08          = in[4*0+i]   | t02;
+       t09          = t07       ^ t05;
+       out[4*1+i]   = t06       ^ t08;
+       out[4*0+i]   =     ~ t09;
+       t12          = in[4*1+i]   & out[4*0+i];
+       t13          = t01       & t05;
+       t14          = t01       ^ t12;
+       t15          = t07       ^ t13;
+       t16          = in[4*3+i]   | t02;
+       t17          = in[4*0+i]   ^ out[4*1+i];
+       out[4*3+i]   = t17       ^ t15;
+       out[4*2+i]   = t16       ^ t14; 
+}
+
+/* S7:   1 13 15  0 14  8  2 11  7  4 12 10  9  3  5  6 */
+
+/* depth = 10,7,10,4, Total gates=19 */
+static void sb7(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t05, t06, t08, t09, t10, t11, t13, t14, t15, t16, t17, t01;
+       t01          = in[4*0+i]   & in[4*2+i];
+       t02          =     ~ in[4*3+i];
+       t03          = in[4*0+i]   & t02;
+       t04          = in[4*1+i]   | t01;
+       t05          = in[4*0+i]   & in[4*1+i];
+       t06          = in[4*2+i]   ^ t04;
+       out[4*3+i]   = t03       ^ t06;
+       t08          = in[4*2+i]   | out[4*3+i];
+       t09          = in[4*3+i]   | t05;
+       t10          = in[4*0+i]   ^ t08;
+       t11          = t04       & out[4*3+i];
+       out[4*1+i]   = t09       ^ t10;
+       t13          = in[4*1+i]   ^ out[4*1+i];
+       t14          = t01       ^ out[4*1+i];
+       t15          = in[4*2+i]   ^ t05;
+       t16          = t11       | t13;
+       t17          = t02       | t14;
+       out[4*0+i]   = t15       ^ t17;
+       out[4*2+i]   = in[4*0+i]   ^ t16; 
+}
+
+/* InvS7:   3  0  6 13  9 14 15  8  5 12 11  7 10  1  4  2 */
+
+/* depth = 9,7,3,3, Total gates=18 */
+static void sb7_inv(uint8_t* out, const uint8_t* in, uint8_t i){
+       uint8_t t02, t03, t04, t06, t07, t08, t09, t10, t11, t13, t14, t15, t16, t01;
+       t01          = in[4*0+i]   & in[4*1+i];
+       t02          = in[4*0+i]   | in[4*1+i];
+       t03          = in[4*2+i]   | t01;
+       t04          = in[4*3+i]   & t02;
+       out[4*3+i]   = t03       ^ t04;
+       t06          = in[4*1+i]   ^ t04;
+       t07          = in[4*3+i]   ^ out[4*3+i];
+       t08          =     ~ t07;
+       t09          = t06       | t08;
+       t10          = in[4*1+i]   ^ in[4*3+i];
+       t11          = in[4*0+i]   | in[4*3+i];
+       out[4*1+i]   = in[4*0+i]   ^ t09;
+       t13          = in[4*2+i]   ^ t06;
+       t14          = in[4*2+i]   & t11;
+       t15          = in[4*3+i]   | out[4*1+i];
+       t16          = t01       | t10;
+       out[4*0+i]   = t13       ^ t15;
+       out[4*2+i]   = t14       ^ t16; 
+}
+
+typedef void(*sb_fpt)(uint8_t*, const uint8_t*, uint8_t i);
+
+sb_fpt sf_tab[] PROGMEM = {
+       sb0, sb1, sb2, sb3, 
+       sb4, sb5, sb6, sb7
+};
+
+sb_fpt sinvf_tab[] PROGMEM = {
+       sb0_inv, sb1_inv, sb2_inv, sb3_inv, 
+       sb4_inv, sb5_inv, sb6_inv, sb7_inv
+};
+
+void sbox128(void * w, uint8_t box){
+       uint8_t i, buffer[16];
+       box &= 0x7;
+       
+       sb_fpt fp;
+       fp = (sb_fpt)pgm_read_word(&(sf_tab[box]));
+       for(i=0; i<4; ++i){
+               fp(buffer, (uint8_t*)w, i);
+       }
+       memcpy(w, buffer, 16);
+}
+
+void inv_sbox128(void * w, uint8_t box){
+       uint8_t i, buffer[16];
+       box &= 0x7;
+       
+       sb_fpt fp;
+       fp = (sb_fpt)pgm_read_word(&(sinvf_tab[box]));
+       for(i=0; i<4; ++i){
+               fp(buffer, (uint8_t*)w, i);
+       }
+       memcpy(w, buffer, 16);
+}
+
+
+
+
+
+
+
+

diff --git a/serpent-sboxes.c b/serpent-sboxes.c
new file mode 100644 (file)
index 0000000..8e54f6a
--- /dev/null
+++ b/serpent-sboxes.c
@@ -0,0 +1,108 @@
+/* serpent-sboxes.c
+ * a non-bitsliced implementation of the serpent sboxes
+ * author: Daniel Otte 
+ * email:  daniel.otte@rub.de
+ * license: GPLv3
+ */
+
+#include <stdint.h>
+#include <string.h> /* memset() */
+#include <avr/pgmspace.h>
+#include "serpent-sboxes.h"
+
+uint8_t sbox[] PROGMEM = {
+ 0x38, 0xF1, 0xA6, 0x5B, 0xED, 0x42, 0x70, 0x9C,
+ 0xFC, 0x27, 0x90, 0x5A, 0x1B, 0xE8, 0x6D, 0x34, 
+ 0x86, 0x79, 0x3C, 0xAF, 0xD1, 0xE4, 0x0B, 0x52,
+ 0x0F, 0xB8, 0xC9, 0x63, 0xD1, 0x24, 0xA7, 0x5E,
+ 0x1F, 0x83, 0xC0, 0xB6, 0x25, 0x4A, 0x9E, 0x7D,
+ 0xF5, 0x2B, 0x4A, 0x9C, 0x03, 0xE8, 0xD6, 0x71,
+ 0x72, 0xC5, 0x84, 0x6B, 0xE9, 0x1F, 0xD3, 0xA0,
+ 0x1D, 0xF0, 0xE8, 0x2B, 0x74, 0xCA, 0x93, 0x56,
+/* now the inverted sboxes */
+ 0xD3, 0xB0, 0xA6, 0x5C, 0x1E, 0x47, 0xF9, 0x82,
+ 0x58, 0x2E, 0xF6, 0xC3, 0xB4, 0x79, 0x1D, 0xA0,
+ 0xC9, 0xF4, 0xBE, 0x12, 0x03, 0x6D, 0x58, 0xA7,
+ 0x09, 0xA7, 0xBE, 0x6D, 0x35, 0xC2, 0x48, 0xF1,
+ 0x50, 0x83, 0xA9, 0x7E, 0x2C, 0xB6, 0x4F, 0xD1,
+ 0x8F, 0x29, 0x41, 0xDE, 0xB6, 0x53, 0x7C, 0xA0,
+ 0xFA, 0x1D, 0x53, 0x60, 0x49, 0xE7, 0x2C, 0x8B,
+ 0x30, 0x6D, 0x9E, 0xF8, 0x5C, 0xB7, 0xA1, 0x42
+};        
+         
+
+#define SHR_O(a) c=(a)&1; ((a) = (a)>>1)
+#define SHR_I(a) ((a) = (c?0x80:0x00)| ((a)>>1))
+
+static void ip(uint8_t *o, uint32_t *i){
+       uint8_t c; // carry 
+       uint8_t n,m;
+       memset(o, 0, 16);
+       for(n=0; n<16; ++n){
+               for(m=0; m<2; ++m){
+               SHR_O(i[0]);
+               SHR_I(o[n]);
+               SHR_O(i[1]);
+               SHR_I(o[n]);
+               SHR_O(i[2]);
+               SHR_I(o[n]);
+               SHR_O(i[3]);
+               SHR_I(o[n]);
+               }
+       }
+}
+
+#undef SHR_I
+#define SHR_I(a) ((a) = (c?0x80000000L:0x00L)| ((a)>>1)) /* we use 32-bit words here */
+
+
+static void fp(uint32_t *o, uint32_t *i){
+       uint8_t c; // carry 
+       uint8_t n,m;
+       memset(o, 0, 16);
+       for(n=0; n<4; ++n){
+               for(m=0; m<8; ++m){
+               SHR_O(i[n]);
+               SHR_I(o[0]);
+               SHR_O(i[n]);
+               SHR_I(o[1]);
+               SHR_O(i[n]);
+               SHR_I(o[2]);
+               SHR_O(i[n]);
+               SHR_I(o[3]);
+               }
+       }
+}
+
+/******************************************************************************/
+
+static void sbox128x(uint8_t box, void* w){
+       uint8_t sb[16];
+       uint8_t i,t,x;
+       box &= 0x0f;
+       /* load sbox */
+       for(i=0; i<8; ++i){
+               t = pgm_read_byte(sbox + box*8 + i);
+               sb[2*i+0]=t>>4;
+               sb[2*i+1]=t&0xf;
+       }
+       uint8_t o[16];
+       ip(o, w);
+       
+       for(i=0; i<16; ++i){
+               t = ((uint8_t*)o)[i];
+               x = sb[t>>4];
+               x <<= 4;
+               x |= sb[t&0xf];
+               ((uint8_t*)o)[i] = x;
+       }
+       fp(w, (uint32_t*)o);
+}
+
+void sbox128(void * w, uint8_t box){
+       sbox128x(box&0x7, w);
+}
+
+void inv_sbox128(void * w, uint8_t box){
+       sbox128x(((box&0x7)|0x8), w);
+}

diff --git a/serpent-sboxes.h b/serpent-sboxes.h
new file mode 100644 (file)
index 0000000..d44309d
--- /dev/null
+++ b/serpent-sboxes.h
@@ -0,0 +1,10 @@
+#ifndef SERPENTSBOXES_H_
+#define SERPENTSBOXES_H_
+
+#include <stdint.h>
+
+void sbox128(void * w, uint8_t box);
+void inv_sbox128(void * w, uint8_t box);
+
+
+#endif /*SERPENTSBOXES_H_*/

diff --git a/serpent.c b/serpent.c
index 85cf4a852e4a731d46fc63e3baf852deeb5acd05..91e5158f38941d2b4959478730e33195c9e06824 100644 (file)
--- a/serpent.c
+++ b/serpent.c
@@ -9,103 +9,7 @@
 #include <string.h> /* memset() */
 #include <avr/pgmspace.h>
 #include "serpent.h"
-
-uint8_t sbox[] PROGMEM = {
- 0x38, 0xF1, 0xA6, 0x5B, 0xED, 0x42, 0x70, 0x9C,
- 0xFC, 0x27, 0x90, 0x5A, 0x1B, 0xE8, 0x6D, 0x34, 
- 0x86, 0x79, 0x3C, 0xAF, 0xD1, 0xE4, 0x0B, 0x52,
- 0x0F, 0xB8, 0xC9, 0x63, 0xD1, 0x24, 0xA7, 0x5E,
- 0x1F, 0x83, 0xC0, 0xB6, 0x25, 0x4A, 0x9E, 0x7D,
- 0xF5, 0x2B, 0x4A, 0x9C, 0x03, 0xE8, 0xD6, 0x71,
- 0x72, 0xC5, 0x84, 0x6B, 0xE9, 0x1F, 0xD3, 0xA0,
- 0x1D, 0xF0, 0xE8, 0x2B, 0x74, 0xCA, 0x93, 0x56,
-/* now the inverted sboxes */
- 0xD3, 0xB0, 0xA6, 0x5C, 0x1E, 0x47, 0xF9, 0x82,
- 0x58, 0x2E, 0xF6, 0xC3, 0xB4, 0x79, 0x1D, 0xA0,
- 0xC9, 0xF4, 0xBE, 0x12, 0x03, 0x6D, 0x58, 0xA7,
- 0x09, 0xA7, 0xBE, 0x6D, 0x35, 0xC2, 0x48, 0xF1,
- 0x50, 0x83, 0xA9, 0x7E, 0x2C, 0xB6, 0x4F, 0xD1,
- 0x8F, 0x29, 0x41, 0xDE, 0xB6, 0x53, 0x7C, 0xA0,
- 0xFA, 0x1D, 0x53, 0x60, 0x49, 0xE7, 0x2C, 0x8B,
- 0x30, 0x6D, 0x9E, 0xF8, 0x5C, 0xB7, 0xA1, 0x42
-};        
-         
-
-#define SHR_O(a) c=(a)&1; ((a) = (a)>>1)
-#define SHR_I(a) ((a) = (c?0x80:0x00)| ((a)>>1))
-
-static void ip(uint8_t *o, uint32_t *i){
-       uint8_t c; // carry 
-       uint8_t n,m;
-       memset(o, 0, 16);
-       for(n=0; n<16; ++n){
-               for(m=0; m<2; ++m){
-               SHR_O(i[0]);
-               SHR_I(o[n]);
-               SHR_O(i[1]);
-               SHR_I(o[n]);
-               SHR_O(i[2]);
-               SHR_I(o[n]);
-               SHR_O(i[3]);
-               SHR_I(o[n]);
-               }
-       }
-}
-
-#undef SHR_I
-#define SHR_I(a) ((a) = (c?0x80000000L:0x00L)| ((a)>>1)) /* we use 32-bit words here */
-
-
-static void fp(uint32_t *o, uint32_t *i){
-       uint8_t c; // carry 
-       uint8_t n,m;
-       memset(o, 0, 16);
-       for(n=0; n<4; ++n){
-               for(m=0; m<8; ++m){
-               SHR_O(i[n]);
-               SHR_I(o[0]);
-               SHR_O(i[n]);
-               SHR_I(o[1]);
-               SHR_O(i[n]);
-               SHR_I(o[2]);
-               SHR_O(i[n]);
-               SHR_I(o[3]);
-               }
-       }
-}
-
-/******************************************************************************/
-
-static void sbox128x(uint8_t box, void* w){
-       uint8_t sb[16];
-       uint8_t i,t,x;
-       box &= 0x0f;
-       /* load sbox */
-       for(i=0; i<8; ++i){
-               t = pgm_read_byte(sbox + box*8 + i);
-               sb[2*i+0]=t>>4;
-               sb[2*i+1]=t&0xf;
-       }
-       uint8_t o[16];
-       ip(o, w);
-       
-       for(i=0; i<16; ++i){
-               t = ((uint8_t*)o)[i];
-               x = sb[t>>4];
-               x <<= 4;
-               x |= sb[t&0xf];
-               ((uint8_t*)o)[i] = x;
-       }
-       fp(w, (uint32_t*)o);
-}
-
-static void sbox128(void * w, uint8_t box){
-       sbox128x(box&0x7, w);
-}
-
-static void inv_sbox128(void * w, uint8_t box){
-       sbox128x(((box&0x7)|0x8), w);
-}
+#include "serpent-sboxes.h"
 
 /******************************************************************************/
 

diff --git a/serpent.mk b/serpent.mk
index eb55908f6909f8cfffb1f23ebef17f8269512588..5975ec0e8604ea177400eb7ff23835f9c54c126c 100644 (file)
--- a/serpent.mk
+++ b/serpent.mk
@@ -5,9 +5,9 @@ ALGO_NAME := SERPENT
 BLOCK_CIPHERS += $(ALGO_NAME)
 
 
-$(ALGO_NAME)_OBJ      := serpent.o
+$(ALGO_NAME)_OBJ      := serpent.o serpent-sboxes-bitslice.o
 $(ALGO_NAME)_TEST_BIN := main-serpent-test.o debug.o uart.o serial-tools.o \
-                         serpent.o nessie_bc_test.o
+                         serpent.o serpent-sboxes-bitslice.o nessie_bc_test.o
 $(ALGO_NAME)_NESSIE_TEST      := "nessie"
 $(ALGO_NAME)_PEROFRMANCE_TEST := "performance"