Lines Matching +full:0 +full:x32
47 // Multiply 32x32 => 64 and accumulate
48 vpshufd $0x10, T0, T4
49 vpshufd $0x32, T0, T0
50 vpshufd $0x10, T1, T5
51 vpshufd $0x32, T1, T1
52 vpshufd $0x10, T2, T6
53 vpshufd $0x32, T2, T2
54 vpshufd $0x10, T3, T7
55 vpshufd $0x32, T3, T3
70 * It's guaranteed that message_len % 16 == 0.
74 vmovdqu 0x00(KEY), K0
75 vmovdqu 0x10(KEY), K1
76 add $0x20, KEY
82 sub $0x40, MESSAGE_LEN
86 vmovdqu 0x00(KEY), K2
87 vmovdqu 0x10(KEY), K3
90 vmovdqu 0x20(MESSAGE), T3
91 vmovdqu 0x20(KEY), K0
92 vmovdqu 0x30(KEY), K1
95 add $0x40, MESSAGE
96 add $0x40, KEY
97 sub $0x40, MESSAGE_LEN
101 and $0x3f, MESSAGE_LEN
104 cmp $0x20, MESSAGE_LEN
109 vmovdqu 0x00(KEY), K2
110 vmovdqu 0x10(KEY), K3
112 add $0x20, MESSAGE
113 add $0x20, KEY
114 sub $0x20, MESSAGE_LEN
124 vmovdqu 0x00(KEY), K2_XMM
125 vmovdqu 0x10(KEY), K3_XMM
131 // PASS0_SUMS is (0A 0B 0C 0D)
136 // (0A + 0B + 0C + 0D,
142 vpunpcklqdq PASS1_SUMS, PASS0_SUMS, T0 // T0 = (0A 1A 0C 1C)
143 vpunpckhqdq PASS1_SUMS, PASS0_SUMS, T1 // T1 = (0B 1B 0D 1D)
147 vinserti128 $0x1, T2_XMM, T0, T4 // T4 = (0A 1A 2A 3A)
148 vinserti128 $0x1, T3_XMM, T1, T5 // T5 = (0B 1B 2B 3B)
149 vperm2i128 $0x31, T2, T0, T0 // T0 = (0C 1C 2C 3C)
150 vperm2i128 $0x31, T3, T1, T1 // T1 = (0D 1D 2D 3D)