media/libvpx/vp8/encoder/arm/armv6/vp8_mse16x16_armv6.asm@b8a032363ba2 (annotated)
media/libvpx/vp8/encoder/arm/armv6/vp8_mse16x16_armv6.asm
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
| michael@0 | 1 | ; |
| michael@0 | 2 | ; Copyright (c) 2011 The WebM project authors. All Rights Reserved. |
| michael@0 | 3 | ; |
| michael@0 | 4 | ; Use of this source code is governed by a BSD-style license |
| michael@0 | 5 | ; that can be found in the LICENSE file in the root of the source |
| michael@0 | 6 | ; tree. An additional intellectual property rights grant can be found |
| michael@0 | 7 | ; in the file PATENTS. All contributing project authors may |
| michael@0 | 8 | ; be found in the AUTHORS file in the root of the source tree. |
| michael@0 | 9 | ; |
| michael@0 | 10 | |
| michael@0 | 11 | |
| michael@0 | 12 | EXPORT |vp8_mse16x16_armv6| |
| michael@0 | 13 | |
| michael@0 | 14 | ARM |
| michael@0 | 15 | |
| michael@0 | 16 | AREA ||.text||, CODE, READONLY, ALIGN=2 |
| michael@0 | 17 | |
| michael@0 | 18 | ; r0 unsigned char *src_ptr |
| michael@0 | 19 | ; r1 int source_stride |
| michael@0 | 20 | ; r2 unsigned char *ref_ptr |
| michael@0 | 21 | ; r3 int recon_stride |
| michael@0 | 22 | ; stack unsigned int *sse |
| michael@0 | 23 | ; |
| michael@0 | 24 | ;note: Based on vp8_variance16x16_armv6. In this function, sum is never used. |
| michael@0 | 25 | ; So, we can remove this part of calculation. |
| michael@0 | 26 | |
| michael@0 | 27 | |vp8_mse16x16_armv6| PROC |
| michael@0 | 28 | |
| michael@0 | 29 | push {r4-r9, lr} |
| michael@0 | 30 | |
| michael@0 | 31 | pld [r0, r1, lsl #0] |
| michael@0 | 32 | pld [r2, r3, lsl #0] |
| michael@0 | 33 | |
| michael@0 | 34 | mov r12, #16 ; set loop counter to 16 (=block height) |
| michael@0 | 35 | mov r4, #0 ; initialize sse = 0 |
| michael@0 | 36 | |
| michael@0 | 37 | loop |
| michael@0 | 38 | ; 1st 4 pixels |
| michael@0 | 39 | ldr r5, [r0, #0x0] ; load 4 src pixels |
| michael@0 | 40 | ldr r6, [r2, #0x0] ; load 4 ref pixels |
| michael@0 | 41 | |
| michael@0 | 42 | mov lr, #0 ; constant zero |
| michael@0 | 43 | |
| michael@0 | 44 | usub8 r8, r5, r6 ; calculate difference |
| michael@0 | 45 | pld [r0, r1, lsl #1] |
| michael@0 | 46 | sel r7, r8, lr ; select bytes with positive difference |
| michael@0 | 47 | usub8 r9, r6, r5 ; calculate difference with reversed operands |
| michael@0 | 48 | pld [r2, r3, lsl #1] |
| michael@0 | 49 | sel r8, r9, lr ; select bytes with negative difference |
| michael@0 | 50 | |
| michael@0 | 51 | ; calculate partial sums |
| michael@0 | 52 | usad8 r5, r7, lr ; calculate sum of positive differences |
| michael@0 | 53 | usad8 r6, r8, lr ; calculate sum of negative differences |
| michael@0 | 54 | orr r8, r8, r7 ; differences of all 4 pixels |
| michael@0 | 55 | |
| michael@0 | 56 | ldr r5, [r0, #0x4] ; load 4 src pixels |
| michael@0 | 57 | |
| michael@0 | 58 | ; calculate sse |
| michael@0 | 59 | uxtb16 r6, r8 ; byte (two pixels) to halfwords |
| michael@0 | 60 | uxtb16 r7, r8, ror #8 ; another two pixels to halfwords |
| michael@0 | 61 | smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1) |
| michael@0 | 62 | |
| michael@0 | 63 | ; 2nd 4 pixels |
| michael@0 | 64 | ldr r6, [r2, #0x4] ; load 4 ref pixels |
| michael@0 | 65 | smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2) |
| michael@0 | 66 | |
| michael@0 | 67 | usub8 r8, r5, r6 ; calculate difference |
| michael@0 | 68 | sel r7, r8, lr ; select bytes with positive difference |
| michael@0 | 69 | usub8 r9, r6, r5 ; calculate difference with reversed operands |
| michael@0 | 70 | sel r8, r9, lr ; select bytes with negative difference |
| michael@0 | 71 | |
| michael@0 | 72 | ; calculate partial sums |
| michael@0 | 73 | usad8 r5, r7, lr ; calculate sum of positive differences |
| michael@0 | 74 | usad8 r6, r8, lr ; calculate sum of negative differences |
| michael@0 | 75 | orr r8, r8, r7 ; differences of all 4 pixels |
| michael@0 | 76 | ldr r5, [r0, #0x8] ; load 4 src pixels |
| michael@0 | 77 | ; calculate sse |
| michael@0 | 78 | uxtb16 r6, r8 ; byte (two pixels) to halfwords |
| michael@0 | 79 | uxtb16 r7, r8, ror #8 ; another two pixels to halfwords |
| michael@0 | 80 | smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1) |
| michael@0 | 81 | |
| michael@0 | 82 | ; 3rd 4 pixels |
| michael@0 | 83 | ldr r6, [r2, #0x8] ; load 4 ref pixels |
| michael@0 | 84 | smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2) |
| michael@0 | 85 | |
| michael@0 | 86 | usub8 r8, r5, r6 ; calculate difference |
| michael@0 | 87 | sel r7, r8, lr ; select bytes with positive difference |
| michael@0 | 88 | usub8 r9, r6, r5 ; calculate difference with reversed operands |
| michael@0 | 89 | sel r8, r9, lr ; select bytes with negative difference |
| michael@0 | 90 | |
| michael@0 | 91 | ; calculate partial sums |
| michael@0 | 92 | usad8 r5, r7, lr ; calculate sum of positive differences |
| michael@0 | 93 | usad8 r6, r8, lr ; calculate sum of negative differences |
| michael@0 | 94 | orr r8, r8, r7 ; differences of all 4 pixels |
| michael@0 | 95 | |
| michael@0 | 96 | ldr r5, [r0, #0xc] ; load 4 src pixels |
| michael@0 | 97 | |
| michael@0 | 98 | ; calculate sse |
| michael@0 | 99 | uxtb16 r6, r8 ; byte (two pixels) to halfwords |
| michael@0 | 100 | uxtb16 r7, r8, ror #8 ; another two pixels to halfwords |
| michael@0 | 101 | smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1) |
| michael@0 | 102 | |
| michael@0 | 103 | ; 4th 4 pixels |
| michael@0 | 104 | ldr r6, [r2, #0xc] ; load 4 ref pixels |
| michael@0 | 105 | smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2) |
| michael@0 | 106 | |
| michael@0 | 107 | usub8 r8, r5, r6 ; calculate difference |
| michael@0 | 108 | add r0, r0, r1 ; set src_ptr to next row |
| michael@0 | 109 | sel r7, r8, lr ; select bytes with positive difference |
| michael@0 | 110 | usub8 r9, r6, r5 ; calculate difference with reversed operands |
| michael@0 | 111 | add r2, r2, r3 ; set dst_ptr to next row |
| michael@0 | 112 | sel r8, r9, lr ; select bytes with negative difference |
| michael@0 | 113 | |
| michael@0 | 114 | ; calculate partial sums |
| michael@0 | 115 | usad8 r5, r7, lr ; calculate sum of positive differences |
| michael@0 | 116 | usad8 r6, r8, lr ; calculate sum of negative differences |
| michael@0 | 117 | orr r8, r8, r7 ; differences of all 4 pixels |
| michael@0 | 118 | |
| michael@0 | 119 | subs r12, r12, #1 ; next row |
| michael@0 | 120 | |
| michael@0 | 121 | ; calculate sse |
| michael@0 | 122 | uxtb16 r6, r8 ; byte (two pixels) to halfwords |
| michael@0 | 123 | uxtb16 r7, r8, ror #8 ; another two pixels to halfwords |
| michael@0 | 124 | smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1) |
| michael@0 | 125 | smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2) |
| michael@0 | 126 | |
| michael@0 | 127 | bne loop |
| michael@0 | 128 | |
| michael@0 | 129 | ; return stuff |
| michael@0 | 130 | ldr r1, [sp, #28] ; get address of sse |
| michael@0 | 131 | mov r0, r4 ; return sse |
| michael@0 | 132 | str r4, [r1] ; store sse |
| michael@0 | 133 | |
| michael@0 | 134 | pop {r4-r9, pc} |
| michael@0 | 135 | |
| michael@0 | 136 | ENDP |
| michael@0 | 137 | |
| michael@0 | 138 | END |
