1 files changed, 64 insertions, 0 deletions
diff --git a/src/Crypto/blake2s-load-sse2.h b/src/Crypto/blake2s-load-sse2.h
new file mode 100644
index 00000000..926eedda
--- /dev/null
+++ b/src/Crypto/blake2s-load-sse2.h
@@ -0,0 +1,64 @@
+/*
+   BLAKE2 reference source code package - optimized C implementations
+
+   Copyright 2012, Samuel Neves <sneves@dei.uc.pt>.  You may use this under the
+   terms of the CC0, the OpenSSL Licence, or the Apache Public License 2.0, at
+   your option.  The terms of these licenses can be found at:
+
+   - CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
+   - OpenSSL license   : https://www.openssl.org/source/license.html
+   - Apache 2.0        : http://www.apache.org/licenses/LICENSE-2.0
+
+   More information about the BLAKE2 hash function can be found at
+   https://blake2.net.
+*/
+
+/* Adapted for VeraCrypt */
+
+
+#ifndef BLAKE2S_LOAD_SSE2_H
+#define BLAKE2S_LOAD_SSE2_H
+
+#define LOAD_MSG_0_1(buf) buf = _mm_set_epi32(m6,m4,m2,m0)
+#define LOAD_MSG_0_2(buf) buf = _mm_set_epi32(m7,m5,m3,m1)
+#define LOAD_MSG_0_3(buf) buf = _mm_set_epi32(m12,m10,m8,m14)
+#define LOAD_MSG_0_4(buf) buf = _mm_set_epi32(m13,m11,m9,m15)
+#define LOAD_MSG_1_1(buf) buf = _mm_set_epi32(m13,m9,m4,m14)
+#define LOAD_MSG_1_2(buf) buf = _mm_set_epi32(m6,m15,m8,m10)
+#define LOAD_MSG_1_3(buf) buf = _mm_set_epi32(m11,m0,m1,m5)
+#define LOAD_MSG_1_4(buf) buf = _mm_set_epi32(m7,m2,m12,m3)
+#define LOAD_MSG_2_1(buf) buf = _mm_set_epi32(m15,m5,m12,m11)
+#define LOAD_MSG_2_2(buf) buf = _mm_set_epi32(m13,m2,m0,m8)
+#define LOAD_MSG_2_3(buf) buf = _mm_set_epi32(m7,m3,m10,m9)
+#define LOAD_MSG_2_4(buf) buf = _mm_set_epi32(m1,m6,m14,m4)
+#define LOAD_MSG_3_1(buf) buf = _mm_set_epi32(m11,m13,m3,m7)
+#define LOAD_MSG_3_2(buf) buf = _mm_set_epi32(m14,m12,m1,m9)
+#define LOAD_MSG_3_3(buf) buf = _mm_set_epi32(m4,m5,m2,m15)
+#define LOAD_MSG_3_4(buf) buf = _mm_set_epi32(m0,m10,m6,m8)
+#define LOAD_MSG_4_1(buf) buf = _mm_set_epi32(m10,m2,m5,m9)
+#define LOAD_MSG_4_2(buf) buf = _mm_set_epi32(m15,m4,m7,m0)
+#define LOAD_MSG_4_3(buf) buf = _mm_set_epi32(m6,m11,m14,m3)
+#define LOAD_MSG_4_4(buf) buf = _mm_set_epi32(m8,m12,m1,m13)
+#define LOAD_MSG_5_1(buf) buf = _mm_set_epi32(m8,m0,m6,m2)
+#define LOAD_MSG_5_2(buf) buf = _mm_set_epi32(m3,m11,m10,m12)
+#define LOAD_MSG_5_3(buf) buf = _mm_set_epi32(m15,m7,m4,m1)
+#define LOAD_MSG_5_4(buf) buf = _mm_set_epi32(m14,m5,m13,m9)
+#define LOAD_MSG_6_1(buf) buf = _mm_set_epi32(m4,m14,m1,m12)
+#define LOAD_MSG_6_2(buf) buf = _mm_set_epi32(m10,m13,m15,m5)
+#define LOAD_MSG_6_3(buf) buf = _mm_set_epi32(m9,m6,m0,m8)
+#define LOAD_MSG_6_4(buf) buf = _mm_set_epi32(m2,m3,m7,m11)
+#define LOAD_MSG_7_1(buf) buf = _mm_set_epi32(m3,m12,m7,m13)
+#define LOAD_MSG_7_2(buf) buf = _mm_set_epi32(m9,m1,m14,m11)
+#define LOAD_MSG_7_3(buf) buf = _mm_set_epi32(m8,m15,m5,m2)
+#define LOAD_MSG_7_4(buf) buf = _mm_set_epi32(m6,m4,m0,m10)
+#define LOAD_MSG_8_1(buf) buf = _mm_set_epi32(m0,m11,m14,m6)
+#define LOAD_MSG_8_2(buf) buf = _mm_set_epi32(m8,m3,m9,m15)
+#define LOAD_MSG_8_3(buf) buf = _mm_set_epi32(m1,m13,m12,m10)
+#define LOAD_MSG_8_4(buf) buf = _mm_set_epi32(m4,m7,m2,m5)
+#define LOAD_MSG_9_1(buf) buf = _mm_set_epi32(m1,m7,m8,m10)
+#define LOAD_MSG_9_2(buf) buf = _mm_set_epi32(m5,m6,m4,m2)
+#define LOAD_MSG_9_3(buf) buf = _mm_set_epi32(m3,m9,m15,m13)
+#define LOAD_MSG_9_4(buf) buf = _mm_set_epi32(m12,m14,m11,m0)
+
+
+#endif