2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
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.
12 EXPORT |vp8_short_walsh4x4_neon|
18 AREA ||.text||, CODE, READONLY, ALIGN=2
19 ;void vp8_short_walsh4x4_neon(short *input, short *output, int pitch)
23 |vp8_short_walsh4x4_neon| PROC
25 vld1.16 {d0}, [r0@64], r2 ; load input
26 vld1.16 {d1}, [r0@64], r2
27 vld1.16 {d2}, [r0@64], r2
31 ;transpose d0, d1, d2, d3. Then, d0=ip[0], d1=ip[1], d2=ip[2], d3=ip[3]
35 vmov.s32 q15, #3 ; add 3 to all values
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]
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
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)
59 ;transpose d0, d1, d2, d3, Then, d0=ip[0], d1=ip[4], d2=ip[8], d3=ip[12]
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]
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
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
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
91 ; vrshrn? would add 1 << 3-1 = 2
97 vst1.16 {q0, q1}, [r1@128]