14 #define b_offset	4
61  * b input register containing b
64  * operations on a and b are interleaved to increase performance
66 #define encrypt_round(a,b,c,d,round)\  argument
67 	movzx	b ## B,		%edi;\
68 	mov	s1(%r11,%rdi,4),%r8d;\
69 	movzx	a ## B,		%edi;\
70 	mov	s2(%r11,%rdi,4),%r9d;\
71 	movzx	b ## H,		%edi;\
72 	ror	$16,		b ## D;\
73 	xor	s2(%r11,%rdi,4),%r8d;\
76 	xor	s3(%r11,%rdi,4),%r9d;\
77 	movzx	b ## B,		%edi;\
78 	xor	s3(%r11,%rdi,4),%r8d;\
79 	movzx	a ## B,		%edi;\
80 	xor	(%r11,%rdi,4),	%r9d;\
81 	movzx	b ## H,		%edi;\
82 	ror	$15,		b ## D;\
83 	xor	(%r11,%rdi,4),	%r8d;\
85 	xor	s1(%r11,%rdi,4),%r9d;\
91 	add	k+4+round(%r11),%r8d;\
96  * b input register containing b
99  * operations on a and b are interleaved to increase performance
100  * during the round a and b are prepared for the output whitening
102 #define encrypt_last_round(a,b,c,d,round)\  argument
103 	mov	b ## D,		%r10d;\
105 	movzx	b ## B,		%edi;\
106 	mov	s1(%r11,%rdi,4),%r8d;\
107 	movzx	a ## B,		%edi;\
108 	mov	s2(%r11,%rdi,4),%r9d;\
109 	movzx	b ## H,		%edi;\
110 	ror	$16,		b ## D;\
111 	xor	s2(%r11,%rdi,4),%r8d;\
114 	xor	s3(%r11,%rdi,4),%r9d;\
115 	movzx	b ## B,		%edi;\
116 	xor	s3(%r11,%rdi,4),%r8d;\
117 	movzx	a ## B,		%edi;\
118 	xor	(%r11,%rdi,4),	%r9d;\
120 	movzx	b ## H,		%edi;\
121 	xor	(%r11,%rdi,4),	%r8d;\
123 	xor	s1(%r11,%rdi,4),%r9d;\
129 	add	k+4+round(%r11),%r8d;\
134  * b input register containing b (rotated 16)
137  * operations on a and b are interleaved to increase performance
139 #define decrypt_round(a,b,c,d,round)\  argument
140 	movzx	a ## B,		%edi;\
141 	mov	(%r11,%rdi,4),	%r9d;\
142 	movzx	b ## B,		%edi;\
143 	mov	s3(%r11,%rdi,4),%r8d;\
146 	xor	s1(%r11,%rdi,4),%r9d;\
147 	movzx	b ## H,		%edi;\
148 	ror	$16,		b ## D;\
149 	xor	(%r11,%rdi,4),	%r8d;\
150 	movzx	a ## B,		%edi;\
151 	xor	s2(%r11,%rdi,4),%r9d;\
152 	movzx	b ## B,		%edi;\
153 	xor	s1(%r11,%rdi,4),%r8d;\
156 	xor	s3(%r11,%rdi,4),%r9d;\
157 	movzx	b ## H,		%edi;\
158 	xor	s2(%r11,%rdi,4),%r8d;\
163 	add	k+4+round(%r11),%r8d;\
169  * b input register containing b
172  * operations on a and b are interleaved to increase performance
173  * during the round a and b are prepared for the output whitening
175 #define decrypt_last_round(a,b,c,d,round)\  argument
176 	movzx	a ## B,		%edi;\
177 	mov	(%r11,%rdi,4),	%r9d;\
178 	movzx	b ## B,		%edi;\
179 	mov	s3(%r11,%rdi,4),%r8d;\
180 	movzx	b ## H,		%edi;\
181 	ror	$16,		b ## D;\
182 	xor	(%r11,%rdi,4),	%r8d;\
184 	mov	b ## D,		%r10d;\
188 	xor	s1(%r11,%rdi,4),%r9d;\
189 	movzx	b ## B,		%edi;\
190 	xor	s1(%r11,%rdi,4),%r8d;\
191 	movzx	a ## B,		%edi;\
192 	xor	s2(%r11,%rdi,4),%r9d;\
193 	movzx	b ## H,		%edi;\
194 	xor	s2(%r11,%rdi,4),%r8d;\
196 	xor	s3(%r11,%rdi,4),%r9d;\
201 	add	k+4+round(%r11),%r8d;\
230 	encrypt_round(R0,R1,R2,R3,4*8);
290 	decrypt_round(R2,R3,R0,R1,4*8);