3 files changed, 30 insertions, 9 deletions
diff --git a/src/libcryptobox/CMakeLists.txt b/src/libcryptobox/CMakeLists.txt
index 491d7f9b5..ecd729d6a 100644
--- a/src/libcryptobox/CMakeLists.txt
+++ b/src/libcryptobox/CMakeLists.txt
@@ -54,6 +54,9 @@ IF(HAVE_SSE2)
 	SET(CHACHASRC ${CHACHASRC} ${CMAKE_CURRENT_SOURCE_DIR}/chacha20/sse2.S)
 	SET(POLYSRC ${POLYSRC} ${CMAKE_CURRENT_SOURCE_DIR}/poly1305/sse2.S)
 ENDIF(HAVE_SSE2)
+IF(HAVE_SSE41)
+	SET(SIPHASHSRC ${SIPHASHSRC} ${CMAKE_CURRENT_SOURCE_DIR}/siphash/sse41.S)
+ENDIF(HAVE_SSE41)
 
 CONFIGURE_FILE(platform_config.h.in platform_config.h)
 INCLUDE_DIRECTORIES("${CMAKE_CURRENT_BINARY_DIR}")
diff --git a/src/libcryptobox/siphash/ref.c b/src/libcryptobox/siphash/ref.c
index 2b20ae34d..1a09f2066 100644
--- a/src/libcryptobox/siphash/ref.c
+++ b/src/libcryptobox/siphash/ref.c
@@ -62,8 +62,8 @@
   } while(0)
 
 
-void
-siphash_ref (uint8_t *out, const uint8_t *in, uint64_t inlen, const uint8_t *k)
+uint64_t
+siphash_ref (const unsigned char k[16], const unsigned char *in, const uint64_t inlen)
 {
 	/* "somepseudorandomlygeneratedbytes" */
 	uint64_t v0 = 0x736f6d6570736575ULL;
@@ -137,7 +137,7 @@ siphash_ref (uint8_t *out, const uint8_t *in, uint64_t inlen, const uint8_t *k)
 		;
 
 	b = v0 ^ v1 ^ v2 ^ v3;
-	U64TO8_LE(out, b);
+	return b;
 
 #ifdef DOUBLE
 	v1 ^= 0xdd;
@@ -146,6 +146,7 @@ siphash_ref (uint8_t *out, const uint8_t *in, uint64_t inlen, const uint8_t *k)
 	for( i=0; i<dROUNDS; ++i ) SIPROUND;
 
 	b = v0 ^ v1 ^ v2 ^ v3;
-	U64TO8_LE( out+8, b );
+
+	return b;
 #endif
 }
diff --git a/src/libcryptobox/siphash/siphash.c b/src/libcryptobox/siphash/siphash.c
index f4c8a1854..f42456b7f 100644
--- a/src/libcryptobox/siphash/siphash.c
+++ b/src/libcryptobox/siphash/siphash.c
@@ -33,11 +33,11 @@ typedef struct siphash_impl_t
 	unsigned long cpu_flags;
 	const char *desc;
 
-	void (*siphash)(uint8_t *out, const uint8_t *in, uint64_t inlen, const uint8_t *k);
+	uint64_t (*siphash) (const unsigned char k[16], const unsigned char *in, const uint64_t inlen);
 } siphash_impl_t;
 
 #define SIPHASH_DECLARE(ext) \
-	void siphash_##ext(uint8_t *out, const uint8_t *in, uint64_t inlen, const uint8_t *k);
+	uint64_t siphash_##ext(const unsigned char k[16], const unsigned char *in, const uint64_t inlen);
 
 #define SIPHASH_IMPL(cpuflags, desc, ext) \
 	{(cpuflags), desc, siphash_##ext}
@@ -45,10 +45,17 @@ typedef struct siphash_impl_t
 
 SIPHASH_DECLARE(ref)
 #define SIPHASH_GENERIC SIPHASH_IMPL(0, "generic", ref)
+#if defined(HAVE_SSE41)
+SIPHASH_DECLARE(sse41)
+#define SIPHASH_SSE41 SIPHASH_IMPL(CPUID_SSE41, "sse41", sse41)
+#endif
 
 /* list implemenations from most optimized to least, with generic as the last entry */
 static const siphash_impl_t siphash_list[] = {
 		SIPHASH_GENERIC,
+#if defined(SIPHASH_SSE41)
+		SIPHASH_SSE41,
+#endif
 };
 
 static const siphash_impl_t *siphash_opt = &siphash_list[0];
@@ -66,17 +73,22 @@ siphash_load(void)
 			}
 		}
 	}
+	fprintf(stderr, "selected %s\n", siphash_opt->desc);
 }
 
 void siphash24 (unsigned char *out, const unsigned char *in,
 		unsigned long long inlen, const unsigned char *k)
 {
-	siphash_opt->siphash (out, in, inlen, k);
+	uint64_t r;
+
+	r = siphash_opt->siphash (k, in, inlen);
+	memcpy (out, &r, sizeof (r));
 }
 
 
 size_t
-siphash24_test (void) {
+siphash24_test (bool generic)
+{
 	static const unsigned char vectors[64][8] = {
 		{ 0x31, 0x0e, 0x0e, 0xdd, 0x47, 0xdb, 0x6f, 0x72, },
 		{ 0xfd, 0x67, 0xdc, 0x93, 0xc5, 0x39, 0xf8, 0x74, },
@@ -159,7 +171,12 @@ siphash24_test (void) {
 		for (i = 0; i < sizeof in; ++i) {
 			in[i] = i;
 
-			siphash24 (r.c, in, i, k);
+			if (generic) {
+				r.m = siphash_list[0].siphash (k, in, i);
+			}
+			else {
+				r.m = siphash_opt->siphash (k, in, i);
+			}
 			if (memcmp (r.c, vectors[i], sizeof (r)) != 0) {
 				return 0;
 			}