X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/0f00dc4c8eb47e67bc0f148c2dd109f73a451e0a..HEAD:/symm/rijndael.c

diff --git a/symm/rijndael.c b/symm/rijndael.c
index 9d8e739f..597ae98f 100644
--- a/symm/rijndael.c
+++ b/symm/rijndael.c
@@ -27,12 +27,20 @@
 
 /*----- Header files ------------------------------------------------------*/
 
+#include "config.h"
+
 #include <assert.h>
 #include <stdio.h>
 
 #include <mLib/bits.h>
 
+#if defined(TEST_RIG) && defined(ENABLE_ASM_DEBUG)
+#  include "regdump.h"
+#  define BLKC_TESTHOOK do { regdump_init(); } while (0)
+#endif
+
 #include "blkc.h"
+#include "dispatch.h"
 #include "gcipher.h"
 #include "rijndael.h"
 #include "rijndael-base.h"
@@ -69,6 +77,69 @@ void rijndael_init(rijndael_ctx *k, const void *buf, size_t sz)
  * Use:		Low-level block encryption and decryption.
  */
 
+CPU_DISPATCH(EMPTY, EMPTY, void, rijndael_eblk,
+	     (const rijndael_ctx *k, const uint32 s[4], uint32 d[4]),
+	     (k, s, d), pick_eblk, simple_eblk)
+
+CPU_DISPATCH(EMPTY, EMPTY, void, rijndael_dblk,
+	     (const rijndael_ctx *k, const uint32 s[4], uint32 d[4]),
+	     (k, s, d), pick_dblk, simple_dblk)
+
+#if CPUFAM_X86 || CPUFAM_AMD64
+extern rijndael_eblk__functype rijndael_eblk_x86ish_aesni;
+extern rijndael_dblk__functype rijndael_dblk_x86ish_aesni;
+extern rijndael_eblk__functype rijndael_eblk_x86ish_aesni_avx;
+extern rijndael_dblk__functype rijndael_dblk_x86ish_aesni_avx;
+#endif
+#if CPUFAM_ARMEL && HAVE_AS_ARMV8_CRYPTO
+extern rijndael_eblk__functype rijndael_eblk_arm_crypto;
+extern rijndael_dblk__functype rijndael_dblk_arm_crypto;
+#endif
+#if CPUFAM_ARM64
+extern rijndael_eblk__functype rijndael_eblk_arm64_crypto;
+extern rijndael_dblk__functype rijndael_dblk_arm64_crypto;
+#endif
+
+static rijndael_eblk__functype *pick_eblk(void)
+{
+#if CPUFAM_X86 || CPUFAM_AMD64
+  DISPATCH_PICK_COND(rijndael_eblk, rijndael_eblk_x86ish_aesni_avx,
+		     cpu_feature_p(CPUFEAT_X86_AVX) &&
+		     cpu_feature_p(CPUFEAT_X86_AESNI));
+  DISPATCH_PICK_COND(rijndael_eblk, rijndael_eblk_x86ish_aesni,
+		     cpu_feature_p(CPUFEAT_X86_AESNI));
+#endif
+#if CPUFAM_ARMEL && HAVE_AS_ARMV8_CRYPTO
+  DISPATCH_PICK_COND(rijndael_eblk, rijndael_eblk_arm_crypto,
+		     cpu_feature_p(CPUFEAT_ARM_AES));
+#endif
+#if CPUFAM_ARM64
+  DISPATCH_PICK_COND(rijndael_eblk, rijndael_eblk_arm64_crypto,
+		     cpu_feature_p(CPUFEAT_ARM_AES));
+#endif
+  DISPATCH_PICK_FALLBACK(rijndael_eblk, simple_eblk);
+}
+
+static rijndael_dblk__functype *pick_dblk(void)
+{
+#if CPUFAM_X86 || CPUFAM_AMD64
+  DISPATCH_PICK_COND(rijndael_dblk, rijndael_dblk_x86ish_aesni_avx,
+		     cpu_feature_p(CPUFEAT_X86_AVX) &&
+		     cpu_feature_p(CPUFEAT_X86_AESNI));
+  DISPATCH_PICK_COND(rijndael_dblk, rijndael_dblk_x86ish_aesni,
+		     cpu_feature_p(CPUFEAT_X86_AESNI));
+#endif
+#if CPUFAM_ARMEL && HAVE_AS_ARMV8_CRYPTO
+  DISPATCH_PICK_COND(rijndael_dblk, rijndael_dblk_arm_crypto,
+		     cpu_feature_p(CPUFEAT_ARM_AES));
+#endif
+#if CPUFAM_ARM64
+  DISPATCH_PICK_COND(rijndael_dblk, rijndael_dblk_arm64_crypto,
+		     cpu_feature_p(CPUFEAT_ARM_AES));
+#endif
+  DISPATCH_PICK_FALLBACK(rijndael_dblk, simple_dblk);
+}
+
 #define DO(what, t, aa, bb, cc, dd, a, b, c, d, w) do {			\
   aa = what(t, a, b, c, d) ^ *w++;					\
   bb = what(t, b, c, d, a) ^ *w++;					\
@@ -83,7 +154,7 @@ void rijndael_init(rijndael_ctx *k, const void *buf, size_t sz)
   dd = what(t, d, c, b, a) ^ *w++;					\
 } while (0)
 
-void rijndael_eblk(const rijndael_ctx *k, const uint32 *s, uint32 *dst)
+static void simple_eblk(const rijndael_ctx *k, const uint32 *s, uint32 *dst)
 {
   uint32 a = s[0], b = s[1], c = s[2], d = s[3];
   uint32 aa, bb, cc, dd;
@@ -118,7 +189,7 @@ void rijndael_eblk(const rijndael_ctx *k, const uint32 *s, uint32 *dst)
   dst[0] = a; dst[1] = b; dst[2] = c; dst[3] = d;
 }
 
-void rijndael_dblk(const rijndael_ctx *k, const uint32 *s, uint32 *dst)
+static void simple_dblk(const rijndael_ctx *k, const uint32 *s, uint32 *dst)
 {
   uint32 a = s[0], b = s[1], c = s[2], d = s[3];
   uint32 aa, bb, cc, dd;