aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/src/asm
diff options
context:
space:
mode:
authorEgor Tensin <Egor.Tensin@gmail.com>2015-06-10 05:13:30 +0300
committerEgor Tensin <Egor.Tensin@gmail.com>2015-06-10 05:13:30 +0300
commit1429346822e56485acc4f4af4d1458e9df4115b9 (patch)
tree5985cca7b2b9877416ec6b227055169014292ae7 /src/asm
parentrename source files (diff)
downloadaes-tools-1429346822e56485acc4f4af4d1458e9df4115b9.tar.gz
aes-tools-1429346822e56485acc4f4af4d1458e9df4115b9.zip
src/aes*.asm -> src/asm/, src/aes*.c -> src/c/
Diffstat (limited to 'src/asm')
-rw-r--r--src/asm/aes128.asm202
-rw-r--r--src/asm/aes192.asm248
-rw-r--r--src/asm/aes256.asm286
3 files changed, 736 insertions, 0 deletions
diff --git a/src/asm/aes128.asm b/src/asm/aes128.asm
new file mode 100644
index 0000000..b49de0e
--- /dev/null
+++ b/src/asm/aes128.asm
@@ -0,0 +1,202 @@
+; Copyright 2015 Egor Tensin <Egor.Tensin@gmail.com>
+; This file is licensed under the terms of the MIT License.
+; See LICENSE.txt for details.
+
+.586
+.xmm
+.model flat
+
+.code
+
+@raw_aes128_encrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesenc xmm0, [ecx + 10h]
+ aesenc xmm0, [ecx + 20h]
+ aesenc xmm0, [ecx + 30h]
+ aesenc xmm0, [ecx + 40h]
+ aesenc xmm0, [ecx + 50h]
+ aesenc xmm0, [ecx + 60h]
+ aesenc xmm0, [ecx + 70h]
+ aesenc xmm0, [ecx + 80h]
+ aesenc xmm0, [ecx + 90h]
+ aesenclast xmm0, [ecx + 0A0h]
+ ret
+@raw_aes128_encrypt_block@20 endp
+
+@raw_aes128_decrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesdec xmm0, [ecx + 10h]
+ aesdec xmm0, [ecx + 20h]
+ aesdec xmm0, [ecx + 30h]
+ aesdec xmm0, [ecx + 40h]
+ aesdec xmm0, [ecx + 50h]
+ aesdec xmm0, [ecx + 60h]
+ aesdec xmm0, [ecx + 70h]
+ aesdec xmm0, [ecx + 80h]
+ aesdec xmm0, [ecx + 90h]
+ aesdeclast xmm0, [ecx + 0A0h]
+ ret
+@raw_aes128_decrypt_block@20 endp
+
+@raw_aes128_expand_key_schedule@20 proc
+ ; A "word" (in terms of the FIPS 187 standard) is a 32-bit block.
+ ; Words are denoted by `w[N]`.
+ ;
+ ; A key schedule is composed of 10 "regular" keys and a dumb key for
+ ; the "whitening" step.
+ ;
+ ; A key schedule is thus composed of 44 "words".
+ ; The FIPS standard includes an algorithm to calculate these words via
+ ; a simple loop:
+ ;
+ ; i = 4
+ ; while i < 44:
+ ; temp = w[i - 1]
+ ; if i % 4 == 0:
+ ; temp = SubWord(RotWord(temp))^Rcon
+ ; w[i] = w[i - 4]^temp
+ ; i = i + 1
+ ;
+ ; The loop above may be unrolled like this:
+ ;
+ ; w[4] = SubWord(RotWord(w[3]))^Rcon^w[0]
+ ; w[5] = w[4]^w[1]
+ ; = SubWord(RotWord(w[3]))^Rcon^w[1]^w[0]
+ ; w[6] = w[5]^w[2]
+ ; = SubWord(RotWord(w[3]))^Rcon^w[2]^w[1]^w[0]
+ ; w[7] = w[6]^w[3]
+ ; = SubWord(RotWord(w[3]))^Rcon^w[3]^w[2]^w[1]^w[0]
+ ; w[8] = SubWord(RotWord(w[7]))^Rcon^w[4]
+ ; w[9] = w[8]^w[5]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[5]^w[4]
+ ; w[10] = w[9]^w[6]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[6]^w[5]^w[4]
+ ; w[11] = w[10]^w[7]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[7]^w[6]^w[5]^w[4]
+ ;
+ ; ... and so on.
+ ;
+ ; The Intel AES-NI instruction set facilitates calculating SubWord
+ ; and RotWord using `aeskeygenassist`, which is used in this routine.
+ ;
+ ; Preconditions:
+ ; * xmm0[127:96] == w[3],
+ ; * xmm0[95:64] == w[2],
+ ; * xmm0[63:32] == w[1],
+ ; * xmm0[31:0] == w[0].
+
+ movdqa [ecx], xmm0 ; sets w[0], w[1], w[2], w[3]
+ add ecx, 10h ; ecx = &w[4]
+
+ aeskeygenassist xmm7, xmm0, 01h ; xmm7[127:96] = RotWord(SubWord(w[3]))^Rcon
+ call aes128_keygen_assist ; sets w[4], w[5], w[6], w[7]
+ aeskeygenassist xmm7, xmm0, 02h ; xmm7[127:96] = RotWord(SubWord(w[7]))^Rcon
+ call aes128_keygen_assist ; sets w[8], w[9], w[10], w[11]
+ aeskeygenassist xmm7, xmm0, 04h ; xmm7[127:96] = RotWord(SubWord(w[11]))^Rcon
+ call aes128_keygen_assist ; sets w[12], w[13], w[14], w[15]
+ aeskeygenassist xmm7, xmm0, 08h ; xmm7[127:96] = RotWord(SubWord(w[15]))^Rcon
+ call aes128_keygen_assist ; sets w[16], w[17], w[18], w[19]
+ aeskeygenassist xmm7, xmm0, 10h ; xmm7[127:96] = RotWord(SubWord(w[19]))^Rcon
+ call aes128_keygen_assist ; sets w[20], w[21], w[22], w[23]
+ aeskeygenassist xmm7, xmm0, 20h ; xmm7[127:96] = RotWord(SubWord(w[23]))^Rcon
+ call aes128_keygen_assist ; sets w[24], w[25], w[26], w[27]
+ aeskeygenassist xmm7, xmm0, 40h ; xmm7[127:96] = RotWord(SubWord(w[27]))^Rcon
+ call aes128_keygen_assist ; sets w[28], w[29], w[30], w[31]
+ aeskeygenassist xmm7, xmm0, 80h ; xmm7[127:96] = RotWord(SubWord(w[31]))^Rcon
+ call aes128_keygen_assist ; sets w[32], w[33], w[34], w[35]
+ aeskeygenassist xmm7, xmm0, 1Bh ; xmm7[127:96] = RotWord(SubWord(w[35]))^Rcon
+ call aes128_keygen_assist ; sets w[36], w[37], w[38], w[39]
+ aeskeygenassist xmm7, xmm0, 36h ; xmm7[127:96] = RotWord(SubWord(w[39]))^Rcon
+ call aes128_keygen_assist ; sets w[40], w[41], w[42], w[43]
+
+ ret
+
+aes128_keygen_assist:
+ ; Preconditions:
+ ; * xmm0[127:96] == w[i+3],
+ ; * xmm0[95:64] == w[i+2],
+ ; * xmm0[63:32] == w[i+1],
+ ; * xmm0[31:0] == w[i],
+ ; * xmm7[127:96] == RotWord(SubWord(w[i+3]))^Rcon,
+ ; * ecx == &w[i+4].
+ ;
+ ; Postconditions:
+ ; * xmm0[127:96] == w[i+7] == RotWord(SubWord(w[i+3]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; * xmm0[95:64] == w[i+6] == RotWord(SubWord(w[i+3]))^Rcon^w[i+2]^w[i+1]^w[i],
+ ; * xmm0[63:32] == w[i+5] == RotWord(SubWord(w[i+3]))^Rcon^w[i+1]^w[i],
+ ; * xmm0[31:0] == w[i+4] == RotWord(SubWord(w[i+3]))^Rcon^w[i],
+ ; * ecx == &w[i+8],
+ ; * the value in xmm6 is also modified.
+
+ ; Calculate
+ ; w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+2]^w[i+1]^w[i],
+ ; w[i+1]^w[i] and
+ ; w[i].
+ movdqa xmm6, xmm0 ; xmm6 = xmm0
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ ; xmm0[127:96] == w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+1]^w[i]
+ ; xmm0[31:0] == w[i]
+
+ ; Calculate
+ ; w[i+7] == RotWord(SubWord(w[i+3]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+6] == RotWord(SubWord(w[i+3]))^Rcon^w[i+2]^w[i+1]^w[i],
+ ; w[i+5] == RotWord(SubWord(w[i+3]))^Rcon^w[i+1]^w[i] and
+ ; w[i+4] == RotWord(SubWord(w[i+3]))^Rcon^w[i].
+ pshufd xmm6, xmm7, 0FFh ; xmm6[127:96] = xmm6[95:64] = xmm6[63:32] = xmm6[31:0] = xmm7[127:96]
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ ; xmm0[127:96] == w[i+7] == RotWord(SubWord(w[i+3]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+6] == RotWord(SubWord(w[i+3]))^Rcon^w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+5] == RotWord(SubWord(w[i+3]))^Rcon^w[i+1]^w[i]
+ ; xmm0[31:0] == w[i+4] == RotWord(SubWord(w[i+3]))^Rcon^w[i]
+
+ ; Set w[i+4], w[i+5], w[i+6] and w[i+7].
+ movdqa [ecx], xmm0 ; w[i+7] = RotWord(SubWord(w[i+3]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; w[i+6] = RotWord(SubWord(w[i+3]))^Rcon^w[i+2]^w[i+1]^w[i]
+ ; w[i+5] = RotWord(SubWord(w[i+3]))^Rcon^w[i+1]^w[i]
+ ; w[i+4] = RotWord(SubWord(w[i+3]))^Rcon^w[i]
+ add ecx, 10h ; ecx = &w[i+8]
+
+ ret
+@raw_aes128_expand_key_schedule@20 endp
+
+@raw_aes128_invert_key_schedule@8 proc
+ movdqa xmm7, [ecx]
+ movdqa xmm6, [ecx + 0A0h]
+ movdqa [edx], xmm6
+ movdqa [edx + 0A0h], xmm7
+
+ aesimc xmm7, [ecx + 10h]
+ aesimc xmm6, [ecx + 90h]
+ movdqa [edx + 10h], xmm6
+ movdqa [edx + 90h], xmm7
+
+ aesimc xmm7, [ecx + 20h]
+ aesimc xmm6, [ecx + 80h]
+ movdqa [edx + 20h], xmm6
+ movdqa [edx + 80h], xmm7
+
+ aesimc xmm7, [ecx + 30h]
+ aesimc xmm6, [ecx + 70h]
+ movdqa [edx + 30h], xmm6
+ movdqa [edx + 70h], xmm7
+
+ aesimc xmm7, [ecx + 40h]
+ aesimc xmm6, [ecx + 60h]
+ movdqa [edx + 40h], xmm6
+ movdqa [edx + 60h], xmm7
+
+ aesimc xmm7, [ecx + 50h]
+ movdqa [edx + 50h], xmm7
+
+ ret
+@raw_aes128_invert_key_schedule@8 endp
+
+end
diff --git a/src/asm/aes192.asm b/src/asm/aes192.asm
new file mode 100644
index 0000000..5cc8ded
--- /dev/null
+++ b/src/asm/aes192.asm
@@ -0,0 +1,248 @@
+; Copyright 2015 Egor Tensin <Egor.Tensin@gmail.com>
+; This file is licensed under the terms of the MIT License.
+; See LICENSE.txt for details.
+
+.586
+.xmm
+.model flat
+
+.code
+
+@raw_aes192_encrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesenc xmm0, [ecx + 10h]
+ aesenc xmm0, [ecx + 20h]
+ aesenc xmm0, [ecx + 30h]
+ aesenc xmm0, [ecx + 40h]
+ aesenc xmm0, [ecx + 50h]
+ aesenc xmm0, [ecx + 60h]
+ aesenc xmm0, [ecx + 70h]
+ aesenc xmm0, [ecx + 80h]
+ aesenc xmm0, [ecx + 90h]
+ aesenc xmm0, [ecx + 0A0h]
+ aesenc xmm0, [ecx + 0B0h]
+ aesenclast xmm0, [ecx + 0C0h]
+ ret
+@raw_aes192_encrypt_block@20 endp
+
+@raw_aes192_decrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesdec xmm0, [ecx + 10h]
+ aesdec xmm0, [ecx + 20h]
+ aesdec xmm0, [ecx + 30h]
+ aesdec xmm0, [ecx + 40h]
+ aesdec xmm0, [ecx + 50h]
+ aesdec xmm0, [ecx + 60h]
+ aesdec xmm0, [ecx + 70h]
+ aesdec xmm0, [ecx + 80h]
+ aesdec xmm0, [ecx + 90h]
+ aesdec xmm0, [ecx + 0A0h]
+ aesdec xmm0, [ecx + 0B0h]
+ aesdeclast xmm0, [ecx + 0C0h]
+ ret
+@raw_aes192_decrypt_block@20 endp
+
+@raw_aes192_expand_key_schedule@36 proc
+ ; A "word" (in terms of the FIPS 187 standard) is a 32-bit block.
+ ; Words are denoted by `w[N]`.
+ ;
+ ; A key schedule is composed of 12 "regular" keys and a dumb key for
+ ; the "whitening" step.
+ ;
+ ; A key schedule is thus composed of 52 "words".
+ ; The FIPS standard includes an algorithm to calculate these words via
+ ; a simple loop:
+ ;
+ ; i = 6
+ ; while i < 52:
+ ; temp = w[i - 1]
+ ; if i % 6 == 0:
+ ; temp = SubWord(RotWord(temp))^Rcon
+ ; w[i] = w[i - 6]^temp
+ ; i = i + 1
+ ;
+ ; The loop above may be unrolled like this:
+ ;
+ ; w[6] = SubWord(RotWord(w[5]))^Rcon^w[0]
+ ; w[7] = w[6]^w[1]
+ ; = SubWord(RotWord(w[5]))^Rcon^w[0]^w[1]
+ ; w[8] = w[7]^w[2]
+ ; = SubWord(RotWord(w[5]))^Rcon^w[0]^w[1]^w[2]
+ ; w[9] = w[8]^w[3]
+ ; = SubWord(RotWord(w[5]))^Rcon^w[0]^w[1]^w[2]^w[3]
+ ; w[10] = w[9]^w[4]
+ ; = SubWord(RotWord(w[5]))^Rcon^w[0]^w[1]^w[2]^w[3]^w[4]
+ ; w[11] = w[10]^w[5]
+ ; = SubWord(RotWord(w[5]))^Rcon^w[0]^w[1]^w[2]^w[3]^w[4]^w[5]
+ ; w[12] = SubWord(RotWord(w[11]))^Rcon^w[6]
+ ; w[13] = w[12]^w[7]
+ ; = SubWord(RotWord(w[11]))^Rcon^w[6]^w[7]
+ ; w[14] = w[13]^w[8]
+ ; = SubWord(RotWord(w[11]))^Rcon^w[6]^w[7]^w[8]
+ ; w[15] = w[14]^w[9]
+ ; = SubWord(RotWord(w[11]))^Rcon^w[6]^w[7]^w[8]^w[9]
+ ; w[16] = w[15]^w[10]
+ ; = SubWord(RotWord(w[11]))^Rcon^w[6]^w[7]^w[8]^w[9]^w[10]
+ ; w[17] = w[16]^w[11]
+ ; = SubWort(RotWord(w[11]))^Rcon^w[6]^w[7]^w[8]^w[9]^w[10]^w[11]
+ ;
+ ; ... and so on.
+ ;
+ ; The Intel AES-NI instruction set facilitates calculating SubWord
+ ; and RotWord using `aeskeygenassist`, which is used in this routine.
+ ;
+ ; Preconditions:
+ ; * xmm1[63:32] == w[5],
+ ; * xmm1[31:0] == w[4],
+ ; * xmm0[127:96] == w[3],
+ ; * xmm0[95:64] == w[2],
+ ; * xmm0[63:32] == w[1],
+ ; * xmm0[31:0] == w[0].
+
+ movdqa [ecx], xmm0 ; sets w[0], w[1], w[2], w[3]
+ movq qword ptr [ecx + 10h], xmm1 ; sets w[4], w[5]
+
+ aeskeygenassist xmm7, xmm1, 1 ; xmm7[63:32] = RotWord(SubWord(w[5]))^Rcon,
+ call aes192_keygen_assist
+ movdqu [ecx + 18h], xmm0
+ movq qword ptr [ecx + 28h], xmm1
+ aeskeygenassist xmm7, xmm1, 2 ; xmm7[63:32] = RotWord(SubWord(w[11]))^Rcon
+ call aes192_keygen_assist
+ movdqa [ecx + 30h], xmm0
+ movq qword ptr [ecx + 40h], xmm1
+ aeskeygenassist xmm7, xmm1, 4 ; xmm7[63:32] = RotWord(SubWord(w[17]))^Rcon
+ call aes192_keygen_assist
+ movdqu [ecx + 48h], xmm0
+ movq qword ptr [ecx + 58h], xmm1
+ aeskeygenassist xmm7, xmm1, 8 ; xmm7[63:32] = RotWord(SubWord(w[23]))^Rcon
+ call aes192_keygen_assist
+ movdqa [ecx + 60h], xmm0
+ movq qword ptr [ecx + 70h], xmm1
+ aeskeygenassist xmm7, xmm1, 10h ; xmm7[63:32] = RotWord(SubWord(w[29]))^Rcon
+ call aes192_keygen_assist
+ movdqu [ecx + 78h], xmm0
+ movq qword ptr [ecx + 88h], xmm1
+ aeskeygenassist xmm7, xmm1, 20h ; xmm7[63:32] = RotWord(SubWord(w[35]))^Rcon
+ call aes192_keygen_assist
+ movdqa [ecx + 90h], xmm0
+ movq qword ptr [ecx + 0a0h], xmm1
+ aeskeygenassist xmm7, xmm1, 40h ; xmm7[63:32] = RotWord(SubWord(w[41]))^Rcon
+ call aes192_keygen_assist
+ movdqu [ecx + 0a8h], xmm0
+ movq qword ptr [ecx + 0b8h], xmm1
+ aeskeygenassist xmm7, xmm1, 80h ; xmm7[63:32] = RotWord(SubWord(w[49]))^Rcon
+ call aes192_keygen_assist
+ movdqa [ecx + 0c0h], xmm0
+
+ ret
+
+aes192_keygen_assist:
+ ; Preconditions:
+ ; * xmm1[127:96] == 0,
+ ; * xmm1[95:64] == 0,
+ ; * xmm1[63:32] == w[i+5],
+ ; * xmm1[31:0] == w[i+4],
+ ; * xmm0[127:96] == w[i+3],
+ ; * xmm0[95:64] == w[i+2],
+ ; * xmm0[63:32] == w[i+1],
+ ; * xmm0[31:0] == w[i],
+ ; * xmm7[63:32] == RotWord(SubWord(w[i+5]))^Rcon.
+ ;
+ ; Postconditions:
+ ; * xmm1[127:96] == 0,
+ ; * xmm1[95:64] == 0,
+ ; * xmm1[63:32] == w[i+11] == RotWord(SubWord(w[i+5]))^Rcon^w[i+5]^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; * xmm1[31:0] == w[i+10] == RotWord(SubWord(w[i+5]))^Rcon^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; * xmm0[127:96] == w[i+9] == RotWord(SubWord(w[i+5]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; * xmm0[95:64] == w[i+8] == RotWord(SubWord(w[i+5]))^Rcon^w[i+2]^w[i+1]^w[i],
+ ; * xmm0[63:32] == w[i+7] == RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i],
+ ; * xmm0[31:0] == w[i+6] == RotWord(SubWord(w[i+5]))^Rcon^w[i],
+ ; * the value in xmm6 is also modified.
+
+ ; Calculate
+ ; w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+2]^w[i+1]^w[i],
+ ; w[i+1]^w[i] and
+ ; w[i].
+ movdqa xmm6, xmm0 ; xmm6 = xmm0
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ ; xmm0[127:96] == w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+1]^w[i]
+ ; xmm0[31:0] == w[i]
+
+ ; Calculate
+ ; w[i+9] == RotWord(SubWord(w[i+5]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+8] == RotWord(SubWord(w[i+5]))^Rcon^w[i+2]^w[i+1]^w[i],
+ ; w[i+7] == RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i] and
+ ; w[i+6] == RotWord(SubWord(w[i+5]))^Rcon^w[i].
+ pshufd xmm6, xmm7, 55h ; xmm6[127:96] = xmm6[95:64] = xmm6[63:32] = xmm6[31:0] = xmm7[63:32]
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ ; xmm0[127:96] == w[i+9] == RotWord(SubWord(w[i+5]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+8] == RotWord(SubWord(w[i+5]))^Rcon^w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+7] == RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i]
+ ; xmm0[31:0] == w[i+6] == RotWord(SubWord(w[i+5]))^Rcon^w[i]
+
+ ; Calculate
+ ; w[i+5]^w[i+4],
+ ; w[i+4].
+ pshufd xmm6, xmm1, 0F3h ; xmm6 = xmm1[31:0] << 32
+ pxor xmm1, xmm6 ; xmm1 ^= xmm7
+ ; xmm1[63:32] == w[i+5]^w[i+4]
+ ; xmm1[31:0] == w[i+4]
+
+ ; Calculate
+ ; w[i+10] == RotWord(SubWord(w[i+5]))^Rcon^w[i+5]^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+11] == RotWord(SubWord(w[i+5]))^Rcon^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i].
+ pshufd xmm6, xmm0, 0FFh ; xmm6[127:96] = xmm6[95:64] = xmm6[63:32] = xmm6[31:0] = xmm0[127:96]
+ psrldq xmm6, 8 ; xmm6 >>= 64
+ pxor xmm1, xmm6 ; xmm1 ^= xmm6
+ ; xmm1[63:32] == w[i+11] == RotWord(SubWord(w[i+5]))^Rcon^w[i+5]^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm1[31:0] == w[i+10] == RotWord(SubWord(w[i+5]))^Rcon^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
+
+ ret
+@raw_aes192_expand_key_schedule@36 endp
+
+@raw_aes192_invert_key_schedule@8 proc
+ movdqa xmm7, [ecx]
+ movdqa xmm6, [ecx + 0C0h]
+ movdqa [edx], xmm6
+ movdqa [edx + 0C0h], xmm7
+
+ aesimc xmm7, [ecx + 10h]
+ aesimc xmm6, [ecx + 0B0h]
+ movdqa [edx + 10h], xmm6
+ movdqa [edx + 0B0h], xmm7
+
+ aesimc xmm7, [ecx + 20h]
+ aesimc xmm6, [ecx + 0A0h]
+ movdqa [edx + 20h], xmm6
+ movdqa [edx + 0A0h], xmm7
+
+ aesimc xmm7, [ecx + 30h]
+ aesimc xmm6, [ecx + 90h]
+ movdqa [edx + 30h], xmm6
+ movdqa [edx + 90h], xmm7
+
+ aesimc xmm7, [ecx + 40h]
+ aesimc xmm6, [ecx + 80h]
+ movdqa [edx + 40h], xmm6
+ movdqa [edx + 80h], xmm7
+
+ aesimc xmm7, [ecx + 50h]
+ aesimc xmm6, [ecx + 70h]
+ movdqa [edx + 50h], xmm6
+ movdqa [edx + 70h], xmm7
+
+ aesimc xmm7, [ecx + 60h]
+ movdqa [edx + 60h], xmm7
+
+ ret
+@raw_aes192_invert_key_schedule@8 endp
+
+end
diff --git a/src/asm/aes256.asm b/src/asm/aes256.asm
new file mode 100644
index 0000000..413e67b
--- /dev/null
+++ b/src/asm/aes256.asm
@@ -0,0 +1,286 @@
+; Copyright 2015 Egor Tensin <Egor.Tensin@gmail.com>
+; This file is licensed under the terms of the MIT License.
+; See LICENSE.txt for details.
+
+.586
+.xmm
+.model flat
+
+.code
+
+@raw_aes256_encrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesenc xmm0, [ecx + 10h]
+ aesenc xmm0, [ecx + 20h]
+ aesenc xmm0, [ecx + 30h]
+ aesenc xmm0, [ecx + 40h]
+ aesenc xmm0, [ecx + 50h]
+ aesenc xmm0, [ecx + 60h]
+ aesenc xmm0, [ecx + 70h]
+ aesenc xmm0, [ecx + 80h]
+ aesenc xmm0, [ecx + 90h]
+ aesenc xmm0, [ecx + 0A0h]
+ aesenc xmm0, [ecx + 0B0h]
+ aesenc xmm0, [ecx + 0C0h]
+ aesenc xmm0, [ecx + 0D0h]
+ aesenclast xmm0, [ecx + 0E0h]
+ ret
+@raw_aes256_encrypt_block@20 endp
+
+@raw_aes256_decrypt_block@20 proc
+ pxor xmm0, [ecx]
+ aesdec xmm0, [ecx + 10h]
+ aesdec xmm0, [ecx + 20h]
+ aesdec xmm0, [ecx + 30h]
+ aesdec xmm0, [ecx + 40h]
+ aesdec xmm0, [ecx + 50h]
+ aesdec xmm0, [ecx + 60h]
+ aesdec xmm0, [ecx + 70h]
+ aesdec xmm0, [ecx + 80h]
+ aesdec xmm0, [ecx + 90h]
+ aesdec xmm0, [ecx + 0A0h]
+ aesdec xmm0, [ecx + 0B0h]
+ aesdec xmm0, [ecx + 0C0h]
+ aesdec xmm0, [ecx + 0D0h]
+ aesdeclast xmm0, [ecx + 0E0h]
+ ret
+@raw_aes256_decrypt_block@20 endp
+
+@raw_aes256_expand_key_schedule@36 proc
+ ; A "word" (in terms of the FIPS 187 standard) is a 32-bit block.
+ ; Words are denoted by `w[N]`.
+ ;
+ ; A key schedule is composed of 14 "regular" keys and a dumb key for
+ ; the "whitening" step.
+ ;
+ ; A key schedule is thus composed of 60 "words".
+ ; The FIPS standard includes an algorithm to calculate these words via
+ ; a simple loop:
+ ;
+ ; i = 8
+ ; while i < 60:
+ ; temp = w[i - 1]
+ ; if i % 8 == 0:
+ ; temp = SubWord(RotWord(temp))^Rcon
+ ; elif i % 8 == 4:
+ ; temp = SubWord(temp)
+ ; w[i] = w[i - 8]^temp
+ ; i = i + 1
+ ;
+ ; The loop above may be unrolled like this:
+ ;
+ ; w[8] = SubWord(RotWord(w[7]))^Rcon^w[0]
+ ; w[9] = w[8]^w[1]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[1]^w[0]
+ ; w[10] = w[9]^w[2]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[2]^w[1]^w[0]
+ ; w[11] = w[10]^w[3]
+ ; = SubWord(RotWord(w[7]))^Rcon^w[3]^w[2]^w[1]^w[0]
+ ; w[12] = SubWord(w[11])^w[4]
+ ; w[13] = w[12]^w[5]
+ ; = SubWord(w[11])^w[5]^w[4]
+ ; w[14] = w[13]^w[6]
+ ; = SubWord(w[11])^w[6]^w[5]^w[4]
+ ; w[15] = w[14]^w[7]
+ ; = SubWord(w[11])^w[7]^w[6]^w[5]^w[4]
+ ; w[16] = SubWord(RotWord(w[15]))^Rcon^w[8]
+ ; w[17] = w[16]^w[9]
+ ; = SubWord(RotWord(w[15]))^Rcon^w[9]^w[8]
+ ; w[18] = w[17]^w[10]
+ ; = SubWord(RotWord(w[15]))^Rcon^w[10]^w[9]^w[8]
+ ; w[19] = w[18]^w[11]
+ ; = SubWord(RotWord(w[15]))^Rcon^w[11]^w[10]^w[9]^w[8]
+ ; w[20] = SubWord(w[19])^w[12]
+ ; w[21] = w[20]^w[13]
+ ; = SubWord(w[19])^w[13]^w[12]
+ ; w[22] = w[21]^w[14]
+ ; = SubWord(w[19])^w[14]^w[13]^w[12]
+ ; w[23] = w[22]^w[15]
+ ; = SubWord(w[19])^w[15]^w[14]^w[13]^w[12]
+ ;
+ ; ... and so on.
+ ;
+ ; The Intel AES-NI instruction set facilitates calculating SubWord
+ ; and RotWord using `aeskeygenassist`, which is used in this routine.
+ ;
+ ; Preconditions:
+ ; * xmm1[127:96] == w[7],
+ ; * xmm1[95:64] == w[6],
+ ; * xmm1[63:32] == w[5],
+ ; * xmm1[31:0] == w[4],
+ ; * xmm0[127:96] == w[3],
+ ; * xmm0[95:64] == w[2],
+ ; * xmm0[63:32] == w[1],
+ ; * xmm0[31:0] == w[0].
+
+ movdqa [ecx], xmm0 ; sets w[0], w[1], w[2], w[3]
+ movdqa [ecx + 10h], xmm1 ; sets w[4], w[5], w[6], w[7]
+ lea ecx, [ecx + 20h] ; ecx = &w[8]
+
+ aeskeygenassist xmm7, xmm1, 1h ; xmm7[127:96] = RotWord(SubWord(w[7]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[8], w[9], w[10], w[11]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[11])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[12], w[13], w[14], w[15]
+
+ aeskeygenassist xmm7, xmm1, 2h ; xmm7[127:96] = RotWord(SubWord(w[15]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[16], w[17], w[18], w[19]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[19])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[20], w[21], w[22], w[23]
+
+ aeskeygenassist xmm7, xmm1, 4h ; xmm7[127:96] = RotWord(SubWord(w[23]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[24], w[25], w[26], w[27]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[27])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[28], w[29], w[30], w[31]
+
+ aeskeygenassist xmm7, xmm1, 8h ; xmm7[127:96] = RotWord(SubWord(w[31]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[32], w[33], w[34], w[35]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[35])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[36], w[37], w[38], w[39]
+
+ aeskeygenassist xmm7, xmm1, 10h ; xmm7[127:96] = RotWord(SubWord(w[39]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[40], w[41], w[42], w[43]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[43])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[44], w[45], w[46], w[47]
+
+ aeskeygenassist xmm7, xmm1, 20h ; xmm7[127:96] = RotWord(SubWord(w[47]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[48], w[49], w[50], w[51]
+
+ aeskeygenassist xmm7, xmm1, 0 ; xmm7[95:64] = SubWord(w[51])
+ pshufd xmm7, xmm7, 0AAh ; xmm7[127:96] = xmm7[63:32] = xmm7[31:0] = xmm7[95:64]
+ call aes256_keygen_assist ; sets w[52], w[53], w[54], w[55]
+
+ aeskeygenassist xmm7, xmm1, 40h ; xmm7[127:96] = RotWord(SubWord(w[55]))^Rcon
+ pshufd xmm7, xmm7, 0FFh ; xmm7[95:64] = xmm7[63:32] = xmm7[31:0] = xmm7[127:96]
+ call aes256_keygen_assist ; sets w[56], w[57], w[58], w[59]
+
+ ret
+
+aes256_keygen_assist:
+ ; Preconditions:
+ ; * xmm1[127:96] == w[i+7],
+ ; * xmm1[95:64] == w[i+6],
+ ; * xmm1[63:32] == w[i+5],
+ ; * xmm1[31:0] == w[i+4],
+ ; * xmm0[127:96] == w[i+3],
+ ; * xmm0[95:64] == w[i+2],
+ ; * xmm0[63:32] == w[i+1],
+ ; * xmm0[31:0] == w[i],
+ ; * xmm7[127:96] == xmm7[95:64] == xmm7[63:32] == xmm7[31:0] == HWGEN,
+ ; where HWGEN is either RotWord(SubWord(w[i+7]))^Rcon or SubWord(w[i+7]),
+ ; depending on the number of the round being processed,
+ ; * ecx == &w[i+8].
+ ;
+ ; Postconditions:
+ ; * xmm1[127:96] == w[i+11] == HWGEN^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; * xmm1[95:64] == w[i+10] == HWGEN^w[i+2]^w[i+1]^w[i],
+ ; * xmm1[63:32] == w[i+9] == HWGEN^w[i+1]^w[i],
+ ; * xmm1[31:0] == w[i+8] == HWGEN^w[i],
+ ; * xmm0[127:96] == w[i+7],
+ ; * xmm0[95:64] == w[i+6],
+ ; * xmm0[63:32] == w[i+5],
+ ; * xmm0[31:0] == w[i+4],
+ ; * ecx == &w[i+12],
+ ; * the value in xmm6 is also modified.
+
+ ; Calculate
+ ; w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; w[i+2]^w[i+1]^w[i],
+ ; w[i+1]^w[i] and
+ ; w[i].
+ movdqa xmm6, xmm0 ; xmm6 = xmm0
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ pslldq xmm6, 4 ; xmm6 <<= 32
+ pxor xmm0, xmm6 ; xmm0 ^= xmm6
+ ; xmm0[127:96] == w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+1]^w[i]
+ ; xmm0[31:0] == w[i]
+
+ ; Calculate
+ ; HWGEN^w[i+3]^w[i+2]^w[i+1]^w[i],
+ ; HWGEN^w[i+2]^w[i+1]^w[i],
+ ; HWGEN^w[i+1]^w[i] and
+ ; HWGEN^w[i].
+ pxor xmm0, xmm7 ; xmm0 ^= xmm7
+ ; xmm0[127:96] == w[i+11] == HWGEN^w[i+3]^w[i+2]^w[i+1]^w[i]
+ ; xmm0[95:64] == w[i+10] == HWGEN^w[i+2]^w[i+1]^w[i]
+ ; xmm0[63:32] == w[i+9] == HWGEN^w[i+1]^w[i]
+ ; xmm0[31:0] == w[i+8] == HWGEN^w[i]
+
+ ; Set w[i+8], w[i+9], w[i+10] and w[i+11].
+ movdqa [ecx], xmm0 ; w[i+8] = HWGEN^w[i]
+ ; w[i+9] = HWGEN^w[i+1]^w[i]
+ ; w[i+10] = HWGEN^w[i+2]^w[i+1]^w[i]
+ ; w[i+11] = HWGEN^w[i+3]^w[i+2]^w[i+1]^w[i]
+ add ecx, 10h ; ecx = &w[i+12]
+
+ ; Swap the values in xmm0 and xmm1.
+ pxor xmm0, xmm1
+ pxor xmm1, xmm0
+ pxor xmm0, xmm1
+
+ ret
+@raw_aes256_expand_key_schedule@36 endp
+
+@raw_aes256_invert_key_schedule@8 proc
+ movdqa xmm7, [ecx]
+ movdqa xmm6, [ecx + 0E0h]
+ movdqa [edx], xmm6
+ movdqa [edx + 0E0h], xmm7
+
+ aesimc xmm7, [ecx + 10h]
+ aesimc xmm6, [ecx + 0D0h]
+ movdqa [edx + 10h], xmm6
+ movdqa [edx + 0D0h], xmm7
+
+ aesimc xmm7, [ecx + 20h]
+ aesimc xmm6, [ecx + 0C0h]
+ movdqa [edx + 20h], xmm6
+ movdqa [edx + 0C0h], xmm7
+
+ aesimc xmm7, [ecx + 30h]
+ aesimc xmm6, [ecx + 0B0h]
+ movdqa [edx + 30h], xmm6
+ movdqa [edx + 0B0h], xmm7
+
+ aesimc xmm7, [ecx + 40h]
+ aesimc xmm6, [ecx + 0A0h]
+ movdqa [edx + 40h], xmm6
+ movdqa [edx + 0A0h], xmm7
+
+ aesimc xmm7, [ecx + 50h]
+ aesimc xmm6, [ecx + 90h]
+ movdqa [edx + 50h], xmm6
+ movdqa [edx + 90h], xmm7
+
+ aesimc xmm7, [ecx + 60h]
+ aesimc xmm6, [ecx + 80h]
+ movdqa [edx + 60h], xmm6
+ movdqa [edx + 80h], xmm7
+
+ aesimc xmm7, [ecx + 70h]
+ movdqa [edx + 70h], xmm7
+
+ ret
+@raw_aes256_invert_key_schedule@8 endp
+
+end