Added AES+GHASH implementation using AES-NI opcodes; also ARM-Thumb assembly for...
[BearSSL] / src / symcipher / aes_x86ni_cbcdec.c
diff --git a/src/symcipher/aes_x86ni_cbcdec.c b/src/symcipher/aes_x86ni_cbcdec.c
new file mode 100644 (file)
index 0000000..6f75348
--- /dev/null
@@ -0,0 +1,224 @@
+/*
+ * Copyright (c) 2017 Thomas Pornin <pornin@bolet.org>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining 
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be 
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "inner.h"
+
+#if BR_AES_X86NI
+
+#if BR_AES_X86NI_GCC
+#include <wmmintrin.h>
+#endif
+
+#if BR_AES_X86NI_MSC
+#include <intrin.h>
+#endif
+
+/* see bearssl_block.h */
+void
+br_aes_x86ni_cbcdec_init(br_aes_x86ni_cbcdec_keys *ctx,
+       const void *key, size_t len)
+{
+       ctx->vtable = &br_aes_x86ni_cbcdec_vtable;
+       ctx->num_rounds = br_aes_x86ni_keysched_dec(ctx->skey.skni, key, len);
+}
+
+/* see bearssl_block.h */
+BR_TARGET("sse2,aes")
+void
+br_aes_x86ni_cbcdec_run(const br_aes_x86ni_cbcdec_keys *ctx,
+       void *iv, void *data, size_t len)
+{
+       unsigned char *buf;
+       unsigned num_rounds;
+       __m128i sk[15], ivx;
+       unsigned u;
+
+       buf = data;
+       ivx = _mm_loadu_si128(iv);
+       num_rounds = ctx->num_rounds;
+       for (u = 0; u <= num_rounds; u ++) {
+               sk[u] = _mm_loadu_si128((void *)(ctx->skey.skni + (u << 4)));
+       }
+       while (len > 0) {
+               __m128i x0, x1, x2, x3, e0, e1, e2, e3;
+
+               x0 = _mm_loadu_si128((void *)(buf +  0));
+               if (len >= 64) {
+                       x1 = _mm_loadu_si128((void *)(buf + 16));
+                       x2 = _mm_loadu_si128((void *)(buf + 32));
+                       x3 = _mm_loadu_si128((void *)(buf + 48));
+               } else {
+                       x0 = _mm_loadu_si128((void *)(buf +  0));
+                       if (len >= 32) {
+                               x1 = _mm_loadu_si128((void *)(buf + 16));
+                               if (len >= 48) {
+                                       x2 = _mm_loadu_si128(
+                                               (void *)(buf + 32));
+                               } else {
+                                       x2 = x0;
+                               }
+                       } else {
+                               x1 = x0;
+                               x2 = x0;
+                       }
+                       x3 = x0;
+               }
+               e0 = x0;
+               e1 = x1;
+               e2 = x2;
+               e3 = x3;
+               x0 = _mm_xor_si128(x0, sk[0]);
+               x1 = _mm_xor_si128(x1, sk[0]);
+               x2 = _mm_xor_si128(x2, sk[0]);
+               x3 = _mm_xor_si128(x3, sk[0]);
+               x0 = _mm_aesdec_si128(x0, sk[1]);
+               x1 = _mm_aesdec_si128(x1, sk[1]);
+               x2 = _mm_aesdec_si128(x2, sk[1]);
+               x3 = _mm_aesdec_si128(x3, sk[1]);
+               x0 = _mm_aesdec_si128(x0, sk[2]);
+               x1 = _mm_aesdec_si128(x1, sk[2]);
+               x2 = _mm_aesdec_si128(x2, sk[2]);
+               x3 = _mm_aesdec_si128(x3, sk[2]);
+               x0 = _mm_aesdec_si128(x0, sk[3]);
+               x1 = _mm_aesdec_si128(x1, sk[3]);
+               x2 = _mm_aesdec_si128(x2, sk[3]);
+               x3 = _mm_aesdec_si128(x3, sk[3]);
+               x0 = _mm_aesdec_si128(x0, sk[4]);
+               x1 = _mm_aesdec_si128(x1, sk[4]);
+               x2 = _mm_aesdec_si128(x2, sk[4]);
+               x3 = _mm_aesdec_si128(x3, sk[4]);
+               x0 = _mm_aesdec_si128(x0, sk[5]);
+               x1 = _mm_aesdec_si128(x1, sk[5]);
+               x2 = _mm_aesdec_si128(x2, sk[5]);
+               x3 = _mm_aesdec_si128(x3, sk[5]);
+               x0 = _mm_aesdec_si128(x0, sk[6]);
+               x1 = _mm_aesdec_si128(x1, sk[6]);
+               x2 = _mm_aesdec_si128(x2, sk[6]);
+               x3 = _mm_aesdec_si128(x3, sk[6]);
+               x0 = _mm_aesdec_si128(x0, sk[7]);
+               x1 = _mm_aesdec_si128(x1, sk[7]);
+               x2 = _mm_aesdec_si128(x2, sk[7]);
+               x3 = _mm_aesdec_si128(x3, sk[7]);
+               x0 = _mm_aesdec_si128(x0, sk[8]);
+               x1 = _mm_aesdec_si128(x1, sk[8]);
+               x2 = _mm_aesdec_si128(x2, sk[8]);
+               x3 = _mm_aesdec_si128(x3, sk[8]);
+               x0 = _mm_aesdec_si128(x0, sk[9]);
+               x1 = _mm_aesdec_si128(x1, sk[9]);
+               x2 = _mm_aesdec_si128(x2, sk[9]);
+               x3 = _mm_aesdec_si128(x3, sk[9]);
+               if (num_rounds == 10) {
+                       x0 = _mm_aesdeclast_si128(x0, sk[10]);
+                       x1 = _mm_aesdeclast_si128(x1, sk[10]);
+                       x2 = _mm_aesdeclast_si128(x2, sk[10]);
+                       x3 = _mm_aesdeclast_si128(x3, sk[10]);
+               } else if (num_rounds == 12) {
+                       x0 = _mm_aesdec_si128(x0, sk[10]);
+                       x1 = _mm_aesdec_si128(x1, sk[10]);
+                       x2 = _mm_aesdec_si128(x2, sk[10]);
+                       x3 = _mm_aesdec_si128(x3, sk[10]);
+                       x0 = _mm_aesdec_si128(x0, sk[11]);
+                       x1 = _mm_aesdec_si128(x1, sk[11]);
+                       x2 = _mm_aesdec_si128(x2, sk[11]);
+                       x3 = _mm_aesdec_si128(x3, sk[11]);
+                       x0 = _mm_aesdeclast_si128(x0, sk[12]);
+                       x1 = _mm_aesdeclast_si128(x1, sk[12]);
+                       x2 = _mm_aesdeclast_si128(x2, sk[12]);
+                       x3 = _mm_aesdeclast_si128(x3, sk[12]);
+               } else {
+                       x0 = _mm_aesdec_si128(x0, sk[10]);
+                       x1 = _mm_aesdec_si128(x1, sk[10]);
+                       x2 = _mm_aesdec_si128(x2, sk[10]);
+                       x3 = _mm_aesdec_si128(x3, sk[10]);
+                       x0 = _mm_aesdec_si128(x0, sk[11]);
+                       x1 = _mm_aesdec_si128(x1, sk[11]);
+                       x2 = _mm_aesdec_si128(x2, sk[11]);
+                       x3 = _mm_aesdec_si128(x3, sk[11]);
+                       x0 = _mm_aesdec_si128(x0, sk[12]);
+                       x1 = _mm_aesdec_si128(x1, sk[12]);
+                       x2 = _mm_aesdec_si128(x2, sk[12]);
+                       x3 = _mm_aesdec_si128(x3, sk[12]);
+                       x0 = _mm_aesdec_si128(x0, sk[13]);
+                       x1 = _mm_aesdec_si128(x1, sk[13]);
+                       x2 = _mm_aesdec_si128(x2, sk[13]);
+                       x3 = _mm_aesdec_si128(x3, sk[13]);
+                       x0 = _mm_aesdeclast_si128(x0, sk[14]);
+                       x1 = _mm_aesdeclast_si128(x1, sk[14]);
+                       x2 = _mm_aesdeclast_si128(x2, sk[14]);
+                       x3 = _mm_aesdeclast_si128(x3, sk[14]);
+               }
+               x0 = _mm_xor_si128(x0, ivx);
+               x1 = _mm_xor_si128(x1, e0);
+               x2 = _mm_xor_si128(x2, e1);
+               x3 = _mm_xor_si128(x3, e2);
+               ivx = e3;
+               _mm_storeu_si128((void *)(buf +  0), x0);
+               if (len >= 64) {
+                       _mm_storeu_si128((void *)(buf + 16), x1);
+                       _mm_storeu_si128((void *)(buf + 32), x2);
+                       _mm_storeu_si128((void *)(buf + 48), x3);
+                       buf += 64;
+                       len -= 64;
+               } else {
+                       if (len >= 32) {
+                               _mm_storeu_si128((void *)(buf + 16), x1);
+                               if (len >= 48) {
+                                       _mm_storeu_si128(
+                                               (void *)(buf + 32), x2);
+                               }
+                       }
+                       break;
+               }
+       }
+       _mm_storeu_si128(iv, ivx);
+}
+
+/* see bearssl_block.h */
+const br_block_cbcdec_class br_aes_x86ni_cbcdec_vtable = {
+       sizeof(br_aes_x86ni_cbcdec_keys),
+       16,
+       4,
+       (void (*)(const br_block_cbcdec_class **, const void *, size_t))
+               &br_aes_x86ni_cbcdec_init,
+       (void (*)(const br_block_cbcdec_class *const *, void *, void *, size_t))
+               &br_aes_x86ni_cbcdec_run
+};
+
+/* see bearssl_block.h */
+const br_block_cbcdec_class *
+br_aes_x86ni_cbcdec_get_vtable(void)
+{
+       return br_aes_x86ni_supported() ? &br_aes_x86ni_cbcdec_vtable : NULL;
+}
+
+#else
+
+/* see bearssl_block.h */
+const br_block_cbcdec_class *
+br_aes_x86ni_cbcdec_get_vtable(void)
+{
+       return NULL;
+}
+
+#endif