1;
2;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3;
4;  Use of this source code is governed by a BSD-style license
5;  that can be found in the LICENSE file in the root of the source
6;  tree. An additional intellectual property rights grant can be found
7;  in the file PATENTS.  All contributing project authors may
8;  be found in the AUTHORS file in the root of the source tree.
9;
10
11
12    EXPORT  |vp8_short_walsh4x4_neon|
13
14    ARM
15    REQUIRE8
16    PRESERVE8
17
18    AREA ||.text||, CODE, READONLY, ALIGN=2
19;void vp8_short_walsh4x4_neon(short *input, short *output, int pitch)
20; r0   short *input,
21; r1   short *output,
22; r2   int pitch
23|vp8_short_walsh4x4_neon| PROC
24
25    vld1.16         {d0}, [r0@64], r2   ; load input
26    vld1.16         {d1}, [r0@64], r2
27    vld1.16         {d2}, [r0@64], r2
28    vld1.16         {d3}, [r0@64]
29
30    ;First for-loop
31    ;transpose d0, d1, d2, d3. Then, d0=ip[0], d1=ip[1], d2=ip[2], d3=ip[3]
32    vtrn.32         d0, d2
33    vtrn.32         d1, d3
34
35    vmov.s32        q15, #3             ; add 3 to all values
36
37    vtrn.16         d0, d1
38    vtrn.16         d2, d3
39
40    vadd.s16        d4, d0, d2          ; ip[0] + ip[2]
41    vadd.s16        d5, d1, d3          ; ip[1] + ip[3]
42    vsub.s16        d6, d1, d3          ; ip[1] - ip[3]
43    vsub.s16        d7, d0, d2          ; ip[0] - ip[2]
44
45    vshl.s16        d4, d4, #2          ; a1 = (ip[0] + ip[2]) << 2
46    vshl.s16        d5, d5, #2          ; d1 = (ip[1] + ip[3]) << 2
47    vshl.s16        d6, d6, #2          ; c1 = (ip[1] - ip[3]) << 2
48    vceq.s16        d16, d4, #0         ; a1 == 0
49    vshl.s16        d7, d7, #2          ; b1 = (ip[0] - ip[2]) << 2
50
51    vadd.s16        d0, d4, d5          ; a1 + d1
52    vmvn            d16, d16            ; a1 != 0
53    vsub.s16        d3, d4, d5          ; op[3] = a1 - d1
54    vadd.s16        d1, d7, d6          ; op[1] = b1 + c1
55    vsub.s16        d2, d7, d6          ; op[2] = b1 - c1
56    vsub.s16        d0, d0, d16         ; op[0] = a1 + d1 + (a1 != 0)
57
58    ;Second for-loop
59    ;transpose d0, d1, d2, d3, Then, d0=ip[0], d1=ip[4], d2=ip[8], d3=ip[12]
60    vtrn.32         d1, d3
61    vtrn.32         d0, d2
62    vtrn.16         d2, d3
63    vtrn.16         d0, d1
64
65    vaddl.s16       q8, d0, d2          ; a1 = ip[0]+ip[8]
66    vaddl.s16       q9, d1, d3          ; d1 = ip[4]+ip[12]
67    vsubl.s16       q10, d1, d3         ; c1 = ip[4]-ip[12]
68    vsubl.s16       q11, d0, d2         ; b1 = ip[0]-ip[8]
69
70    vadd.s32        q0, q8, q9          ; a2 = a1 + d1
71    vadd.s32        q1, q11, q10        ; b2 = b1 + c1
72    vsub.s32        q2, q11, q10        ; c2 = b1 - c1
73    vsub.s32        q3, q8, q9          ; d2 = a1 - d1
74
75    vclt.s32        q8, q0, #0
76    vclt.s32        q9, q1, #0
77    vclt.s32        q10, q2, #0
78    vclt.s32        q11, q3, #0
79
80    ; subtract -1 (or 0)
81    vsub.s32        q0, q0, q8          ; a2 += a2 < 0
82    vsub.s32        q1, q1, q9          ; b2 += b2 < 0
83    vsub.s32        q2, q2, q10         ; c2 += c2 < 0
84    vsub.s32        q3, q3, q11         ; d2 += d2 < 0
85
86    vadd.s32        q8, q0, q15         ; a2 + 3
87    vadd.s32        q9, q1, q15         ; b2 + 3
88    vadd.s32        q10, q2, q15        ; c2 + 3
89    vadd.s32        q11, q3, q15        ; d2 + 3
90
91    ; vrshrn? would add 1 << 3-1 = 2
92    vshrn.s32       d0, q8, #3
93    vshrn.s32       d1, q9, #3
94    vshrn.s32       d2, q10, #3
95    vshrn.s32       d3, q11, #3
96
97    vst1.16         {q0, q1}, [r1@128]
98
99    bx              lr
100
101    ENDP
102
103    END
104