Réutilisation des multiplications dans l'implémentation FELICS

Plutôt que  d'ajouter -Wno-unused-functin, on pourrait  aussi déclarer
les multiplications "static inline", ce qui par convention esquive les
avertissements sur la non-utilisation.

1 files changed, 7 insertions, 104 deletions

diff --git a/src/add_felicsref/tweakey.c b/src/add_felicsref/tweakey.c
index 635c179..c5532f8 100644
--- a/src/add_felicsref/tweakey.c
+++ b/src/add_felicsref/tweakey.c
@@ -15,30 +15,25 @@ http://creativecommons.org/publicdomain/zero/1.0/
 ---
 
 This file provides an implementation of Lilliput-TBC's tweakey schedule,
-similar to the reference implementation save for a few manual optimizations:
+similar to the reference implementation save for one manual optimization:
+the loop over an array of function pointers was unrolled.
 
-- unused multiplication functions were removed using preprocessor
-  conditionals based on the number of lanes;
-
-- the loop over an array of function pointers was unrolled.
-
-These handmade optimizations have been found to significantly decrease code
+This handmade optimization has been found to significantly decrease code
 size and execution time on GCC versions used in the FELICS framework.
 
-This suggests that the compiler does not detect dead code nor does it
-recognize unrolling opportunities, despite the multiplication functions
-being static and thus limited in scope to the compilation unit.
+This suggests that the compiler does not recognize inlining opportunities,
+despite the multiplication functions being static and thus limited in scope
+to the compilation unit.
 */
 
 #include <stdint.h>
 #include <string.h>
 
 #include "constants.h"
+#include "multiplications.h"
 #include "tweakey.h"
 
 
-#define LANE_BITS  64
-#define LANE_BYTES (LANE_BITS/8)
 #define LANES_NB   (TWEAKEY_BYTES/LANE_BYTES)
 
 
@@ -75,98 +70,6 @@ void tweakey_state_extract(
 }
 
 
-static void _multiply_M(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    y[7] = x[6];
-    y[6] = x[5];
-    y[5] = x[5]<<3 ^ x[4];
-    y[4] = x[4]>>3 ^ x[3];
-    y[3] = x[2];
-    y[2] = x[6]<<2 ^ x[1];
-    y[1] = x[0];
-    y[0] = x[7];
-}
-
-static void _multiply_M2(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    uint8_t x_M_5 = x[5]<<3 ^ x[4];
-    uint8_t x_M_4 = x[4]>>3 ^ x[3];
-
-    y[7] = x[5];
-    y[6] = x_M_5;
-    y[5] = x_M_5<<3 ^ x_M_4;
-    y[4] = x_M_4>>3 ^ x[2];
-    y[3] = x[6]<<2  ^ x[1];
-    y[2] = x[5]<<2  ^ x[0];
-    y[1] = x[7];
-    y[0] = x[6];
-}
-
-static void _multiply_M3(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    uint8_t x_M_5  = x[5]<<3  ^ x[4];
-    uint8_t x_M_4  = x[4]>>3  ^ x[3];
-    uint8_t x_M2_5 = x_M_5<<3 ^ x_M_4;
-    uint8_t x_M2_4 = x_M_4>>3 ^ x[2];
-
-    y[7] = x_M_5;
-    y[6] = x_M2_5;
-    y[5] = x_M2_5<<3 ^ x_M2_4;
-    y[4] = x_M2_4>>3 ^ x[6]<<2 ^ x[1];
-    y[3] = x[5]<<2   ^ x[0];
-    y[2] = x_M_5<<2  ^ x[7];
-    y[1] = x[6];
-    y[0] = x[5];
-}
-
-#if LANES_NB >= 5
-static void _multiply_MR(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    y[0] = x[1];
-    y[1] = x[2];
-    y[2] = x[3]    ^ x[4]>>3;
-    y[3] = x[4];
-    y[4] = x[5]    ^ x[6]<<3;
-    y[5] = x[3]<<2 ^ x[6];
-    y[6] = x[7];
-    y[7] = x[0];
-}
-
-#if LANES_NB >= 6
-static void _multiply_MR2(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    uint8_t x_MR_4 = x[5] ^ x[6]<<3;
-
-    y[0] = x[2];
-    y[1] = x[3]    ^ x[4]>>3;
-    y[2] = x[4]    ^ x_MR_4>>3;
-    y[3] = x_MR_4;
-    y[4] = x[3]<<2 ^ x[6]      ^ x[7]<<3;
-    y[5] = x[4]<<2 ^ x[7];
-    y[6] = x[0];
-    y[7] = x[1];
-}
-
-#if LANES_NB >= 7
-static void _multiply_MR3(const uint8_t x[LANE_BYTES], uint8_t y[LANE_BYTES])
-{
-    uint8_t x_MR_4  = x[5]    ^ x[6]<<3;
-    uint8_t x_MR2_4 = x[3]<<2 ^ x[6]    ^ x[7]<<3;
-
-    y[0] = x[3]      ^ x[4]>>3;
-    y[1] = x[4]      ^ x_MR_4>>3;
-    y[2] = x_MR_4    ^ x_MR2_4>>3;
-    y[3] = x_MR2_4;
-    y[4] = x[0]<<3   ^ x[4]<<2   ^ x[7];
-    y[5] = x_MR_4<<2 ^ x[0];
-    y[6] = x[1];
-    y[7] = x[2];
-}
-#endif
-#endif
-#endif
-
-
 void tweakey_state_update(uint8_t TK[TWEAKEY_BYTES])
 {
     /* Skip lane 0, as it is multiplied by the identity matrix. */